This PR enables transitively pinning (TP) objects from particular roots for Immix/StickyImmix (#897)

- A new bucket `ImmovableClosure` has been created to trace and
transitively pin roots, i.e., roots in which no object in its transitive
closure is allowed to move.
- `RootsWorkFactory` has two new functions
`create_process_tp_edge_roots_work` and
`create_process_tp_node_roots_work` to create work to process these
roots.
- `GCWorkContext` expects an `TPProcessEdges` type, which performs the
transitively pinning trace (and is unimplemented for unsupported plans).
- `create_process_node_roots_work` creates work to process roots in the
`NodeRootsTrace` bucket, which is executed after `TPClosure`, but any
work derived from it is put into the regular `Closure` bucket, meaning
that the binding shouldn't need to pin root nodes.
- For sticky immix, currently we only support non-moving nursery
collections (`sticky_immix_non_moving_nursery`), but the solution here
is to use the `TPClosure` on the modified buffer constructed from the
write barrier (since during nursery collection we do not know if these
objects should be transitively pinned or not).

---------

Co-authored-by: Yi Lin <[email protected]>
Co-authored-by: Kunshan Wang <[email protected]>

Author: 3 people

.github/scripts/replace-mmtk-dep.py

@@ -19,10 +19,15 @@
 
 mmtk_node = toml_data["dependencies"]["mmtk"]
 
-print("Deleting dependencies.mmtk.git")
-if "git" in mmtk_node:
-    del mmtk_node["git"]
-
+# These keys may specify the locations of the dependency. Remove them.
+for key in ["git", "branch", "version", "registry"]:
+    if key in mmtk_node:
+        print("Deleting dependencies.mmtk.{}".format(key))
+        del mmtk_node[key]
+    else:
+        print("Key dependencies.mmtk.{} does not exist.  Ignored.".format(key))
+
+# Use mmtk-core from the specified local directory.
 mmtk_repo_path = os.path.realpath(args.mmtk_core_path)
 print("Setting dependencies.mmtk.path to {}".format(mmtk_repo_path))
 mmtk_node["path"] = mmtk_repo_path

Cargo.toml

@@ -105,6 +105,12 @@ object_pinning = []
 # Disable any object copying in Immix. This makes Immix a non-moving policy.
 immix_non_moving = []
 
+# Disable any object copying in nursery GC for Sticky Immix while allowing other kinds of copying.
+# `immix_non_moving` disables all kinds of copying in Immix, so this feature is not needed
+# if `immix_non_moving` is in use.
+sticky_immix_non_moving_nursery = []
+
+
 # Reduce block size for ImmixSpace.  This mitigates fragmentation when defrag is disabled.
 immix_smaller_block = []
 # Zero the unmarked lines after a GC cycle in immix. This helps debug untraced objects.

docs/userguide/src/tutorial/code/mygc_semispace/gc_work.rs

@@ -1,6 +1,6 @@
 // ANCHOR: imports
 use super::global::MyGC;
-use crate::scheduler::gc_work::*;
+use crate::scheduler::{gc_work::*, WorkBucketStage};
 use crate::vm::VMBinding;
 use std::ops::{Deref, DerefMut};
 // ANCHOR_END: imports
@@ -11,24 +11,26 @@ impl<VM: VMBinding> crate::scheduler::GCWorkContext for MyGCWorkContext<VM> {
     type VM = VM;
     type PlanType = MyGC<VM>;
     type ProcessEdgesWorkType = SFTProcessEdges<Self::VM>;
+    type TPProcessEdges = UnsupportedProcessEdges<Self::VM>;
 }
 // ANCHOR_END: workcontext_sft
 
 // ANCHOR: workcontext_plan
-use crate::scheduler::gc_work::PlanProcessEdges;
 use crate::policy::gc_work::DEFAULT_TRACE;
+use crate::scheduler::gc_work::PlanProcessEdges;
 pub struct MyGCWorkContext2<VM: VMBinding>(std::marker::PhantomData<VM>);
 impl<VM: VMBinding> crate::scheduler::GCWorkContext for MyGCWorkContext2<VM> {
     type VM = VM;
     type PlanType = MyGC<VM>;
     type ProcessEdgesWorkType = PlanProcessEdges<Self::VM, MyGC<VM>, DEFAULT_TRACE>;
+    type TPProcessEdges = UnsupportedProcessEdges<Self::VM>;
 }
 // ANCHOR_END: workcontext_plan
 
-use crate::util::ObjectReference;
+use crate::policy::space::Space;
 use crate::util::copy::CopySemantics;
+use crate::util::ObjectReference;
 use crate::MMTK;
-use crate::policy::space::Space;
 
 // ANCHOR: mygc_process_edges
 pub struct MyGCProcessEdges<VM: VMBinding> {
@@ -38,12 +40,17 @@ pub struct MyGCProcessEdges<VM: VMBinding> {
 // ANCHOR_END: mygc_process_edges
 
 // ANCHOR: mygc_process_edges_impl
-impl<VM:VMBinding> ProcessEdgesWork for MyGCProcessEdges<VM> {
+impl<VM: VMBinding> ProcessEdgesWork for MyGCProcessEdges<VM> {
     type VM = VM;
     type ScanObjectsWorkType = ScanObjects<Self>;
 
-    fn new(edges: Vec<EdgeOf<Self>>, roots: bool, mmtk: &'static MMTK<VM>) -> Self {
-        let base = ProcessEdgesBase::new(edges, roots, mmtk);
+    fn new(
+        edges: Vec<EdgeOf<Self>>,
+        roots: bool,
+        mmtk: &'static MMTK<VM>,
+        bucket: WorkBucketStage,
+    ) -> Self {
+        let base = ProcessEdgesBase::new(edges, roots, mmtk, bucket);
         let plan = base.plan().downcast_ref::<MyGC<VM>>().unwrap();
         Self { base, plan }
     }
@@ -74,7 +81,7 @@ impl<VM:VMBinding> ProcessEdgesWork for MyGCProcessEdges<VM> {
     }
 
     fn create_scan_work(&self, nodes: Vec<ObjectReference>, roots: bool) -> ScanObjects<Self> {
-        ScanObjects::<Self>::new(nodes, false, roots)
+        ScanObjects::<Self>::new(nodes, false, roots, self.bucket)
     }
 }
 // ANCHOR_END: mygc_process_edges_impl
@@ -100,5 +107,6 @@ impl<VM: VMBinding> crate::scheduler::GCWorkContext for MyGCWorkContext3<VM> {
     type VM = VM;
     type PlanType = MyGC<VM>;
     type ProcessEdgesWorkType = MyGCProcessEdges<Self::VM>;
+    type TPProcessEdges = UnsupportedProcessEdges<Self::VM>;
 }
 // ANCHOR: workcontext_mygc

src/plan/generational/copying/gc_work.rs

@@ -3,18 +3,20 @@ use crate::plan::generational::gc_work::GenNurseryProcessEdges;
 use crate::vm::*;
 
 use crate::policy::gc_work::DEFAULT_TRACE;
-use crate::scheduler::gc_work::PlanProcessEdges;
+use crate::scheduler::gc_work::{PlanProcessEdges, UnsupportedProcessEdges};
 
 pub struct GenCopyNurseryGCWorkContext<VM: VMBinding>(std::marker::PhantomData<VM>);
 impl<VM: VMBinding> crate::scheduler::GCWorkContext for GenCopyNurseryGCWorkContext<VM> {
     type VM = VM;
     type PlanType = GenCopy<VM>;
     type ProcessEdgesWorkType = GenNurseryProcessEdges<Self::VM, Self::PlanType>;
+    type TPProcessEdges = UnsupportedProcessEdges<VM>;
 }
 
 pub struct GenCopyGCWorkContext<VM: VMBinding>(std::marker::PhantomData<VM>);
 impl<VM: VMBinding> crate::scheduler::GCWorkContext for GenCopyGCWorkContext<VM> {
     type VM = VM;
     type PlanType = GenCopy<VM>;
     type ProcessEdgesWorkType = PlanProcessEdges<Self::VM, GenCopy<VM>, DEFAULT_TRACE>;
+    type TPProcessEdges = UnsupportedProcessEdges<VM>;
 }

src/plan/generational/gc_work.rs

@@ -1,7 +1,7 @@
 use atomic::Ordering;
 
 use crate::plan::PlanTraceObject;
-use crate::scheduler::{gc_work::*, GCWork, GCWorker};
+use crate::scheduler::{gc_work::*, GCWork, GCWorker, WorkBucketStage};
 use crate::util::ObjectReference;
 use crate::vm::edge_shape::{Edge, MemorySlice};
 use crate::vm::*;
@@ -25,8 +25,13 @@ impl<VM: VMBinding, P: GenerationalPlanExt<VM> + PlanTraceObject<VM>> ProcessEdg
     type VM = VM;
     type ScanObjectsWorkType = PlanScanObjects<Self, P>;
 
-    fn new(edges: Vec<EdgeOf<Self>>, roots: bool, mmtk: &'static MMTK<VM>) -> Self {
-        let base = ProcessEdgesBase::new(edges, roots, mmtk);
+    fn new(
+        edges: Vec<EdgeOf<Self>>,
+        roots: bool,
+        mmtk: &'static MMTK<VM>,
+        bucket: WorkBucketStage,
+    ) -> Self {
+        let base = ProcessEdgesBase::new(edges, roots, mmtk, bucket);
         let plan = base.plan().downcast_ref().unwrap();
         Self { plan, base }
     }
@@ -51,7 +56,7 @@ impl<VM: VMBinding, P: GenerationalPlanExt<VM> + PlanTraceObject<VM>> ProcessEdg
         nodes: Vec<ObjectReference>,
         roots: bool,
     ) -> Self::ScanObjectsWorkType {
-        PlanScanObjects::new(self.plan, nodes, false, roots)
+        PlanScanObjects::new(self.plan, nodes, false, roots, self.bucket)
     }
 }
 
@@ -111,7 +116,7 @@ impl<E: ProcessEdgesWork> GCWork<E::VM> for ProcessModBuf<E> {
             // Scan objects in the modbuf and forward pointers
             let modbuf = std::mem::take(&mut self.modbuf);
             GCWork::do_work(
-                &mut ScanObjects::<E>::new(modbuf, false, false),
+                &mut ScanObjects::<E>::new(modbuf, false, false, WorkBucketStage::Closure),
                 worker,
                 mmtk,
             )
@@ -154,7 +159,11 @@ impl<E: ProcessEdgesWork> GCWork<E::VM> for ProcessRegionModBuf<E> {
                 }
             }
             // Forward entries
-            GCWork::do_work(&mut E::new(edges, false, mmtk), worker, mmtk)
+            GCWork::do_work(
+                &mut E::new(edges, false, mmtk, WorkBucketStage::Closure),
+                worker,
+                mmtk,
+            )
         }
     }
 }

src/plan/generational/immix/gc_work.rs

@@ -2,13 +2,15 @@ use super::global::GenImmix;
 use crate::plan::generational::gc_work::GenNurseryProcessEdges;
 use crate::policy::gc_work::TraceKind;
 use crate::scheduler::gc_work::PlanProcessEdges;
+use crate::scheduler::gc_work::UnsupportedProcessEdges;
 use crate::vm::VMBinding;
 
 pub struct GenImmixNurseryGCWorkContext<VM: VMBinding>(std::marker::PhantomData<VM>);
 impl<VM: VMBinding> crate::scheduler::GCWorkContext for GenImmixNurseryGCWorkContext<VM> {
     type VM = VM;
     type PlanType = GenImmix<VM>;
     type ProcessEdgesWorkType = GenNurseryProcessEdges<VM, Self::PlanType>;
+    type TPProcessEdges = UnsupportedProcessEdges<VM>;
 }
 
 pub(super) struct GenImmixMatureGCWorkContext<VM: VMBinding, const KIND: TraceKind>(
@@ -20,4 +22,5 @@ impl<VM: VMBinding, const KIND: TraceKind> crate::scheduler::GCWorkContext
     type VM = VM;
     type PlanType = GenImmix<VM>;
     type ProcessEdgesWorkType = PlanProcessEdges<VM, GenImmix<VM>, KIND>;
+    type TPProcessEdges = UnsupportedProcessEdges<VM>;
 }

src/plan/immix/gc_work.rs

@@ -1,5 +1,6 @@
 use super::global::Immix;
 use crate::policy::gc_work::TraceKind;
+use crate::policy::gc_work::TRACE_KIND_TRANSITIVE_PIN;
 use crate::scheduler::gc_work::PlanProcessEdges;
 use crate::vm::VMBinding;
 
@@ -12,4 +13,5 @@ impl<VM: VMBinding, const KIND: TraceKind> crate::scheduler::GCWorkContext
     type VM = VM;
     type PlanType = Immix<VM>;
     type ProcessEdgesWorkType = PlanProcessEdges<VM, Immix<VM>, KIND>;
+    type TPProcessEdges = PlanProcessEdges<VM, Immix<VM>, TRACE_KIND_TRANSITIVE_PIN>;
 }

src/plan/immix/global.rs

@@ -157,8 +157,8 @@ impl<VM: VMBinding> Immix<VM> {
     /// to schedule a full heap collection. A plan must call set_collection_kind and set_gc_status before this method.
     pub(crate) fn schedule_immix_full_heap_collection<
         PlanType: Plan<VM = VM>,
-        FastContext: 'static + GCWorkContext<VM = VM, PlanType = PlanType>,
-        DefragContext: 'static + GCWorkContext<VM = VM, PlanType = PlanType>,
+        FastContext: GCWorkContext<VM = VM, PlanType = PlanType>,
+        DefragContext: GCWorkContext<VM = VM, PlanType = PlanType>,
     >(
         plan: &'static DefragContext::PlanType,
         immix_space: &ImmixSpace<VM>,

src/plan/markcompact/gc_work.rs

@@ -57,12 +57,13 @@ impl<VM: VMBinding> GCWork<VM> for UpdateReferences<VM> {
             .get_and_clear_worker_live_bytes();
 
         for mutator in VM::VMActivePlan::mutators() {
-            mmtk.scheduler.work_buckets[WorkBucketStage::SecondRoots]
-                .add(ScanMutatorRoots::<ForwardingProcessEdges<VM>>(mutator));
+            mmtk.scheduler.work_buckets[WorkBucketStage::SecondRoots].add(ScanMutatorRoots::<
+                MarkCompactForwardingGCWorkContext<VM>,
+            >(mutator));
         }
 
         mmtk.scheduler.work_buckets[WorkBucketStage::SecondRoots]
-            .add(ScanVMSpecificRoots::<ForwardingProcessEdges<VM>>::new());
+            .add(ScanVMSpecificRoots::<MarkCompactForwardingGCWorkContext<VM>>::new());
     }
 }
 
@@ -102,4 +103,13 @@ impl<VM: VMBinding> crate::scheduler::GCWorkContext for MarkCompactGCWorkContext
     type VM = VM;
     type PlanType = MarkCompact<VM>;
     type ProcessEdgesWorkType = MarkingProcessEdges<VM>;
+    type TPProcessEdges = UnsupportedProcessEdges<VM>;
+}
+
+pub struct MarkCompactForwardingGCWorkContext<VM: VMBinding>(std::marker::PhantomData<VM>);
+impl<VM: VMBinding> crate::scheduler::GCWorkContext for MarkCompactForwardingGCWorkContext<VM> {
+    type VM = VM;
+    type PlanType = MarkCompact<VM>;
+    type ProcessEdgesWorkType = ForwardingProcessEdges<VM>;
+    type TPProcessEdges = UnsupportedProcessEdges<VM>;
 }

src/plan/markcompact/global.rs

@@ -92,7 +92,7 @@ impl<VM: VMBinding> Plan for MarkCompact<VM> {
 
         // Stop & scan mutators (mutator scanning can happen before STW)
         scheduler.work_buckets[WorkBucketStage::Unconstrained]
-            .add(StopMutators::<MarkingProcessEdges<VM>>::new());
+            .add(StopMutators::<MarkCompactGCWorkContext<VM>>::new());
 
         // Prepare global/collectors/mutators
         scheduler.work_buckets[WorkBucketStage::Prepare]