ada: Implement per-finalization-collection spinlocks

This changes the implementation of finalization collections from using the
global task lock to using per-collection spinlocks.  Spinlocks are a good
fit in this context because they are very cheap and therefore can be taken
with a fine granularity only around the portions of code implementing the
shuffling of pointers required by attachment and detachment actions.

gcc/ada/

	* libgnat/s-finpri.ads (Lock_Type): New modular type.
	(Collection_Node): Add Enclosing_Collection component.
	(Finalization_Collection): Add Lock component.
	* libgnat/s-finpri.adb: Add clauses for System.Atomic_Primitives.
	(Attach_Object_To_Collection): Lock and unlock the collection.
	Save a pointer to the enclosing collection in the node.
	(Detach_Object_From_Collection): Lock and unlock the collection.
	(Finalize): Likewise.
	(Initialize): Initialize the lock.
	(Lock_Collection): New procedure.
	(Unlock_Collection): Likewise.

0 files changed, 0 insertions, 0 deletions