[Sablevm-developer] Question about _svmf_exit_object_monitor()

[Archie C. <ar...@de...>]

Etienne-

Question about _svmf_exit_object_monitor(). In the case where the lock
is fat and the thread reference count is going to zero (so that the
monitor is really being released), and there is contention, you iterate
through the list of other threads contending for the monitor and wake
them up. But if one of these threads is actually contending for a
different object monitor (owned by the current thread) than the one
we're presently unlocking, then you inflate that object's lock.

My question is: is this inflate operation strictly necessary? Won't
these other threads properly handle inflating the lock when they wake up?

This question has the following implications: inflating a lock can
possibly throw an exception if we run out of memory. This is in fact
the only failure mode of _svmf_exit_object_monitor() other than throwing
an IllegalMonitorStateException.

For internal VM locks, we can assume that IllegalMonitorStateException
will never be thrown unless there is a bug. So if we can also eliminate
the call to _svmf_inflate_lock_no_exception(), then we can assume that
_svmf_exit_object_monitor() will never return an error except in case
of a bug. This would then make using object locks internally to the VM
easier, because we'd never have a weird failure case to handle when unlocking.

Thanks,
-Archie

__________________________________________________________________________
Archie Cobbs     *     Precision I/O      *     http://www.precisionio.com