2 files changed, 12 insertions, 6 deletions
diff --git a/rts/sm/NonMoving.c b/rts/sm/NonMoving.c
index 41510e7f8e..130da95c3e 100644
--- a/rts/sm/NonMoving.c
+++ b/rts/sm/NonMoving.c
@@ -85,6 +85,10 @@ Mutex concurrent_coll_finished_lock;
  *  - A set of *filled* segments, which contain no unallocated blocks and will
  *    be collected during the next major GC cycle
  *
+ * These sets are maintained as atomic singly-linked lists. This is not
+ * susceptible to the ABA problem since we are guaranteed to push a given
+ * segment to a list only once per garbage collection cycle.
+ *
  * Storage for segments is allocated using the block allocator using an aligned
  * group of NONMOVING_SEGMENT_BLOCKS blocks. This makes the task of locating
  * the segment header for a clone a simple matter of bit-masking (as
@@ -543,7 +547,7 @@ void nonmovingPushFreeSegment(struct NonmovingSegment *seg)
 static struct NonmovingSegment *nonmovingPopFreeSegment(void)
 {
     while (true) {
-        struct NonmovingSegment *seg = nonmovingHeap.free;
+        struct NonmovingSegment *seg = ACQUIRE_LOAD(&nonmovingHeap.free);
         if (seg == NULL) {
             return NULL;
         }
@@ -641,13 +645,15 @@ static bool advance_next_free(struct NonmovingSegment *seg, const unsigned int b
 static struct NonmovingSegment *pop_active_segment(struct NonmovingAllocator *alloca)
 {
     while (true) {
-        struct NonmovingSegment *seg = alloca->active;
+        // Synchronizes with CAS in nonmovingPushActiveSegment
+        struct NonmovingSegment *seg = ACQUIRE_LOAD(&alloca->active);
         if (seg == NULL) {
             return NULL;
         }
+        struct NonmovingSegment *next = RELAXED_LOAD(&seg->link);
         if (cas((StgVolatilePtr) &alloca->active,
                 (StgWord) seg,
-                (StgWord) seg->link) == (StgWord) seg) {
+                (StgWord) next) == (StgWord) seg) {
             return seg;
         }
     }
diff --git a/rts/sm/NonMoving.h b/rts/sm/NonMoving.h
index 12fb9ddaab..1d40ef726e 100644
--- a/rts/sm/NonMoving.h
+++ b/rts/sm/NonMoving.h
@@ -169,7 +169,7 @@ INLINE_HEADER void nonmovingPushActiveSegment(struct NonmovingSegment *seg)
         nonmovingHeap.allocators[nonmovingSegmentLogBlockSize(seg) - NONMOVING_ALLOCA0];
     SET_SEGMENT_STATE(seg, ACTIVE);
     while (true) {
-        struct NonmovingSegment *current_active = (struct NonmovingSegment*)VOLATILE_LOAD(&alloc->active);
+        struct NonmovingSegment *current_active = RELAXED_LOAD(&alloc->active);
         seg->link = current_active;
         if (cas((StgVolatilePtr) &alloc->active, (StgWord) current_active, (StgWord) seg) == (StgWord) current_active) {
             break;
@@ -184,8 +184,8 @@ INLINE_HEADER void nonmovingPushFilledSegment(struct NonmovingSegment *seg)
         nonmovingHeap.allocators[nonmovingSegmentLogBlockSize(seg) - NONMOVING_ALLOCA0];
     SET_SEGMENT_STATE(seg, FILLED);
     while (true) {
-        struct NonmovingSegment *current_filled = (struct NonmovingSegment*)VOLATILE_LOAD(&alloc->filled);
-        seg->link = current_filled;
+        struct NonmovingSegment *current_filled = (struct NonmovingSegment*) RELAXED_LOAD(&alloc->filled);
+        RELAXED_STORE(&seg->link, current_filled);
         if (cas((StgVolatilePtr) &alloc->filled, (StgWord) current_filled, (StgWord) seg) == (StgWord) current_filled) {
             break;
         }