Choose Non-moving Policy based on features (#1308)

This PR uses an Immix space as the default non moving space, and adds
two features to use mark sweep or immortal as the non moving space.
* `Mutator` now includes 2 Immix allocators (one for Immix as the
default space, and the other for Immix as the non moving space).
* Call prepare/release/end_of_gc for the non moving space.
* Add `common_prepare_func` and `common_release_func` for mutators.

Author: qinsoon

Cargo.toml

@@ -224,14 +224,17 @@ malloc_jemalloc = ["dep:jemalloc-sys"]
 # is not compiled in default builds.
 malloc_native_mimalloc = []
 
-# If there are more groups, they should be inserted above this line
-# Group:end
-
 # Group:marksweepallocation
 # default is native allocator with lazy sweeping
 eager_sweeping = []
 # Use library malloc as the freelist allocator for mark sweep. This will makes mark sweep slower. As malloc may return addresses outside our
 # normal heap range, we will have to use chunk-based SFT table. Turning on this feature will use a different SFT map implementation on 64bits,
 # and will affect all the plans in the build. Please be aware of the consequence, and this is only meant to be experimental use.
 malloc_mark_sweep = []
+
+# Group:nonmovingspace
+immortal_as_nonmoving = []
+marksweep_as_nonmoving = []
+
+# If there are more groups, they should be inserted above this line
 # Group:end

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

@@ -128,6 +128,12 @@ impl<VM: VMBinding> Plan for MyGC<VM> {
     }
     // ANCHOR_END: release
 
+    // ANCHOR: end_of_gc
+    fn end_of_gc(&mut self, tls: VMWorkerThread) {
+        self.common.end_of_gc(tls);
+    }
+    // ANCHOR_END: end_of_gc
+
     // Modify
     // ANCHOR: plan_get_collection_reserve
     fn get_collection_reserved_pages(&self) -> usize {

docs/userguide/src/tutorial/mygc/ss/collection.md

@@ -157,6 +157,14 @@ with `&mygc_mutator_release`.  This function will be called at the release stage
 allocation semantics to the new tospace. When the mutator threads resume, any new allocations for
 `Default` will then go to the new tospace.
 
+### End of GC
+
+Find the method `end_of_gc()` in `mygc/global.rs`. Call `end_of_gc` from the common plan instead.
+
+```rust
+{{#include ../../code/mygc_semispace/global.rs:end_of_gc}}
+```
+
 ## ProcessEdgesWork for MyGC
 
 [`ProcessEdgesWork`](https://docs.mmtk.io/api/mmtk/scheduler/gc_work/trait.ProcessEdgesWork.html)

src/plan/generational/copying/global.rs

@@ -114,9 +114,9 @@ impl<VM: VMBinding> Plan for GenCopy<VM> {
         }
     }
 
-    fn end_of_gc(&mut self, _tls: VMWorkerThread) {
-        self.gen
-            .set_next_gc_full_heap(CommonGenPlan::should_next_gc_be_full_heap(self));
+    fn end_of_gc(&mut self, tls: VMWorkerThread) {
+        let next_gc_full_heap = CommonGenPlan::should_next_gc_be_full_heap(self);
+        self.gen.end_of_gc(tls, next_gc_full_heap);
     }
 
     fn get_collection_reserved_pages(&self) -> usize {

src/plan/generational/copying/mutator.rs

@@ -3,7 +3,8 @@ use super::GenCopy;
 use crate::plan::barriers::ObjectBarrier;
 use crate::plan::generational::barrier::GenObjectBarrierSemantics;
 use crate::plan::generational::create_gen_space_mapping;
-use crate::plan::mutator_context::unreachable_prepare_func;
+use crate::plan::mutator_context::common_prepare_func;
+use crate::plan::mutator_context::common_release_func;
 use crate::plan::mutator_context::Mutator;
 use crate::plan::mutator_context::MutatorBuilder;
 use crate::plan::mutator_context::MutatorConfig;
@@ -13,7 +14,7 @@ use crate::util::{VMMutatorThread, VMWorkerThread};
 use crate::vm::VMBinding;
 use crate::MMTK;
 
-pub fn gencopy_mutator_release<VM: VMBinding>(mutator: &mut Mutator<VM>, _tls: VMWorkerThread) {
+pub fn gencopy_mutator_release<VM: VMBinding>(mutator: &mut Mutator<VM>, tls: VMWorkerThread) {
     // reset nursery allocator
     let bump_allocator = unsafe {
         mutator
@@ -23,6 +24,8 @@ pub fn gencopy_mutator_release<VM: VMBinding>(mutator: &mut Mutator<VM>, _tls: V
     .downcast_mut::<BumpAllocator<VM>>()
     .unwrap();
     bump_allocator.reset();
+
+    common_release_func(mutator, tls);
 }
 
 pub fn create_gencopy_mutator<VM: VMBinding>(
@@ -36,7 +39,7 @@ pub fn create_gencopy_mutator<VM: VMBinding>(
             mmtk.get_plan(),
             &gencopy.gen.nursery,
         )),
-        prepare_func: &unreachable_prepare_func,
+        prepare_func: &common_prepare_func,
         release_func: &gencopy_mutator_release,
     };
 

src/plan/generational/global.rs

@@ -78,6 +78,11 @@ impl<VM: VMBinding> CommonGenPlan<VM> {
         self.nursery.release();
     }
 
+    pub fn end_of_gc(&mut self, tls: VMWorkerThread, next_gc_full_heap: bool) {
+        self.set_next_gc_full_heap(next_gc_full_heap);
+        self.common.end_of_gc(tls);
+    }
+
     /// Independent of how many pages remain in the page budget (a function of heap size), we must
     /// ensure we never exhaust virtual memory. Therefore we must never let the nursery grow to the
     /// extent that it can't be copied into the mature space.

src/plan/generational/immix/global.rs

@@ -131,7 +131,7 @@ impl<VM: VMBinding> Plan for GenImmix<VM> {
         if full_heap {
             self.immix_space.prepare(
                 full_heap,
-                crate::policy::immix::defrag::StatsForDefrag::new(self),
+                Some(crate::policy::immix::defrag::StatsForDefrag::new(self)),
             );
         }
     }
@@ -146,9 +146,9 @@ impl<VM: VMBinding> Plan for GenImmix<VM> {
             .store(full_heap, Ordering::Relaxed);
     }
 
-    fn end_of_gc(&mut self, _tls: VMWorkerThread) {
-        self.gen
-            .set_next_gc_full_heap(CommonGenPlan::should_next_gc_be_full_heap(self));
+    fn end_of_gc(&mut self, tls: VMWorkerThread) {
+        let next_gc_full_heap = CommonGenPlan::should_next_gc_be_full_heap(self);
+        self.gen.end_of_gc(tls, next_gc_full_heap);
 
         let did_defrag = self.immix_space.end_of_gc();
         self.last_gc_was_defrag.store(did_defrag, Ordering::Relaxed);

src/plan/generational/immix/mutator.rs

@@ -3,7 +3,8 @@ use crate::plan::barriers::ObjectBarrier;
 use crate::plan::generational::barrier::GenObjectBarrierSemantics;
 use crate::plan::generational::create_gen_space_mapping;
 use crate::plan::generational::immix::GenImmix;
-use crate::plan::mutator_context::unreachable_prepare_func;
+use crate::plan::mutator_context::common_prepare_func;
+use crate::plan::mutator_context::common_release_func;
 use crate::plan::mutator_context::Mutator;
 use crate::plan::mutator_context::MutatorBuilder;
 use crate::plan::mutator_context::MutatorConfig;
@@ -13,7 +14,7 @@ use crate::util::{VMMutatorThread, VMWorkerThread};
 use crate::vm::VMBinding;
 use crate::MMTK;
 
-pub fn genimmix_mutator_release<VM: VMBinding>(mutator: &mut Mutator<VM>, _tls: VMWorkerThread) {
+pub fn genimmix_mutator_release<VM: VMBinding>(mutator: &mut Mutator<VM>, tls: VMWorkerThread) {
     // reset nursery allocator
     let bump_allocator = unsafe {
         mutator
@@ -23,6 +24,8 @@ pub fn genimmix_mutator_release<VM: VMBinding>(mutator: &mut Mutator<VM>, _tls:
     .downcast_mut::<BumpAllocator<VM>>()
     .unwrap();
     bump_allocator.reset();
+
+    common_release_func(mutator, tls);
 }
 
 pub fn create_genimmix_mutator<VM: VMBinding>(
@@ -36,7 +39,7 @@ pub fn create_genimmix_mutator<VM: VMBinding>(
             mmtk.get_plan(),
             &genimmix.gen.nursery,
         )),
-        prepare_func: &unreachable_prepare_func,
+        prepare_func: &common_prepare_func,
         release_func: &genimmix_mutator_release,
     };
 

src/plan/generational/mod.rs

@@ -50,7 +50,6 @@ pub const GEN_CONSTRAINTS: PlanConstraints = PlanConstraints {
     may_trace_duplicate_edges: ACTIVE_BARRIER.equals(BarrierSelector::ObjectBarrier),
     max_non_los_default_alloc_bytes:
         crate::plan::plan_constraints::MAX_NON_LOS_ALLOC_BYTES_COPYING_PLAN,
-    needs_prepare_mutator: false,
     ..PlanConstraints::default()
 };
 

src/plan/global.rs

@@ -197,7 +197,7 @@ pub trait Plan: 'static + HasSpaces + Sync + Downcast {
 
     /// Inform the plan about the end of a GC. It is guaranteed that there is no further work for this GC.
     /// This is invoked once per GC by one worker thread. `tls` is the worker thread that executes this method.
-    fn end_of_gc(&mut self, _tls: VMWorkerThread) {}
+    fn end_of_gc(&mut self, _tls: VMWorkerThread);
 
     /// Notify the plan that an emergency collection will happen. The plan should try to free as much memory as possible.
     /// The default implementation will force a full heap collection for generational plans.
@@ -511,6 +511,10 @@ impl<VM: VMBinding> BasePlan<VM> {
         self.vm_space.release();
     }
 
+    pub fn end_of_gc(&mut self, _tls: VMWorkerThread) {
+        // Do nothing here. None of the spaces needs end_of_gc.
+    }
+
     pub(crate) fn collection_required<P: Plan>(&self, plan: &P, space_full: bool) -> bool {
         let stress_force_gc =
             crate::util::heap::gc_trigger::GCTrigger::<VM>::should_do_stress_gc_inner(
@@ -542,6 +546,17 @@ impl<VM: VMBinding> BasePlan<VM> {
     }
 }
 
+cfg_if::cfg_if! {
+    // Use immortal or mark sweep as the non moving space if the features are enabled. Otherwise use Immix.
+    if #[cfg(feature = "immortal_as_nonmoving")] {
+        pub type NonMovingSpace<VM> = crate::policy::immortalspace::ImmortalSpace<VM>;
+    } else if #[cfg(feature = "marksweep_as_nonmoving")] {
+        pub type NonMovingSpace<VM> = crate::policy::marksweepspace::native_ms::MarkSweepSpace<VM>;
+    } else {
+        pub type NonMovingSpace<VM> = crate::policy::immix::ImmixSpace<VM>;
+    }
+}
+
 /**
 CommonPlan is for representing state and features used by _many_ plans, but that are not fundamental to _all_ plans.  Examples include the Large Object Space and an Immortal space.  Features that are fundamental to _all_ plans must be included in BasePlan.
 */
@@ -551,9 +566,12 @@ pub struct CommonPlan<VM: VMBinding> {
     pub immortal: ImmortalSpace<VM>,
     #[space]
     pub los: LargeObjectSpace<VM>,
-    // TODO: We should use a marksweep space for nonmoving.
     #[space]
-    pub nonmoving: ImmortalSpace<VM>,
+    #[cfg_attr(
+        not(any(feature = "immortal_as_nonmoving", feature = "marksweep_as_nonmoving")),
+        post_scan
+    )] // Immix space needs post_scan
+    pub nonmoving: NonMovingSpace<VM>,
     #[parent]
     pub base: BasePlan<VM>,
 }
@@ -571,12 +589,7 @@ impl<VM: VMBinding> CommonPlan<VM> {
                 args.get_space_args("los", true, false, VMRequest::discontiguous()),
                 false,
             ),
-            nonmoving: ImmortalSpace::new(args.get_space_args(
-                "nonmoving",
-                true,
-                false,
-                VMRequest::discontiguous(),
-            )),
+            nonmoving: Self::new_nonmoving_space(&mut args),
             base: BasePlan::new(args),
         }
     }
@@ -591,17 +604,22 @@ impl<VM: VMBinding> CommonPlan<VM> {
     pub fn prepare(&mut self, tls: VMWorkerThread, full_heap: bool) {
         self.immortal.prepare();
         self.los.prepare(full_heap);
-        self.nonmoving.prepare();
+        self.prepare_nonmoving_space(full_heap);
         self.base.prepare(tls, full_heap)
     }
 
     pub fn release(&mut self, tls: VMWorkerThread, full_heap: bool) {
         self.immortal.release();
         self.los.release(full_heap);
-        self.nonmoving.release();
+        self.release_nonmoving_space(full_heap);
         self.base.release(tls, full_heap)
     }
 
+    pub fn end_of_gc(&mut self, tls: VMWorkerThread) {
+        self.end_of_gc_nonmoving_space();
+        self.base.end_of_gc(tls);
+    }
+
     pub fn get_immortal(&self) -> &ImmortalSpace<VM> {
         &self.immortal
     }
@@ -610,9 +628,65 @@ impl<VM: VMBinding> CommonPlan<VM> {
         &self.los
     }
 
-    pub fn get_nonmoving(&self) -> &ImmortalSpace<VM> {
+    pub fn get_nonmoving(&self) -> &NonMovingSpace<VM> {
         &self.nonmoving
     }
+
+    fn new_nonmoving_space(args: &mut CreateSpecificPlanArgs<VM>) -> NonMovingSpace<VM> {
+        let space_args = args.get_space_args("nonmoving", true, false, VMRequest::discontiguous());
+        cfg_if::cfg_if! {
+            if #[cfg(any(feature = "immortal_as_nonmoving", feature = "marksweep_as_nonmoving"))] {
+                NonMovingSpace::new(space_args)
+            } else {
+                // Immix requires extra args.
+                NonMovingSpace::new(
+                    space_args,
+                    crate::policy::immix::ImmixSpaceArgs {
+                        unlog_object_when_traced: false,
+                        #[cfg(feature = "vo_bit")]
+                        mixed_age: false,
+                        never_move_objects: true,
+                    },
+                )
+            }
+        }
+    }
+
+    fn prepare_nonmoving_space(&mut self, _full_heap: bool) {
+        cfg_if::cfg_if! {
+            if #[cfg(feature = "immortal_as_nonmoving")] {
+                self.nonmoving.prepare();
+            } else if #[cfg(feature = "marksweep_as_nonmoving")] {
+                self.nonmoving.prepare(_full_heap);
+            } else {
+                self.nonmoving.prepare(_full_heap, None);
+            }
+        }
+    }
+
+    fn release_nonmoving_space(&mut self, _full_heap: bool) {
+        cfg_if::cfg_if! {
+            if #[cfg(feature = "immortal_as_nonmoving")] {
+                self.nonmoving.release();
+            } else if #[cfg(feature = "marksweep_as_nonmoving")] {
+                self.nonmoving.prepare(_full_heap);
+            } else {
+                self.nonmoving.release(_full_heap);
+            }
+        }
+    }
+
+    fn end_of_gc_nonmoving_space(&mut self) {
+        cfg_if::cfg_if! {
+            if #[cfg(feature = "immortal_as_nonmoving")] {
+                // Nothing we need to do for immortal space.
+            } else if #[cfg(feature = "marksweep_as_nonmoving")] {
+                self.nonmoving.end_of_gc();
+            } else {
+                self.nonmoving.end_of_gc();
+            }
+        }
+    }
 }
 
 use crate::policy::gc_work::TraceKind;