No scanning an object twice in each GC.

This PR forces each object to be scanned at most once in each GC.  In
this way, no two GC workers will scan the same object at the same time.
This can solve some potential problems such as data races in cleaning
the code cache.  Currently this only affects the MarkSweep plan.

Author: wks

mmtk/Cargo.lock

@@ -37,7 +37,7 @@ features = ["is_mmtk_object", "object_pinning", "sticky_immix_non_moving_nursery
 
 # Uncomment the following lines to use mmtk-core from the official repository.
 git = "https://github.com/mmtk/mmtk-core.git"
-rev = "ec745353a8de72b645613e0fef3ab7f5f1ad9bd1"
+rev = "68bf1b638263b250b12e55ef25bf8d09b01ca0b0"
 
 # Uncomment the following line to use mmtk-core from a local repository.
 #path = "../../mmtk-core"

mmtk/Cargo.toml

@@ -10,6 +10,11 @@ use mmtk::{Mutator, MutatorContext};
 pub struct VMScanning {}
 
 impl Scanning<Ruby> for VMScanning {
+    /// CRuby may do some non-thread-safe operations, such as cleaning up the call cache, while
+    /// scanning an object.  We force each object to be scanned at most once during each GC.  This
+    /// currently only affects the MarkSweep plan.
+    const UNIQUE_OBJECT_ENQUEUING: bool = true;
+
     fn support_slot_enqueuing(_tls: VMWorkerThread, _object: ObjectReference) -> bool {
         false
     }