path: root/rts/sm/Evac.c

/* -----------------------------------------------------------------------------
 *
 * (c) The GHC Team 1998-2008
 *
 * Generational garbage collector: evacuation functions
 *
 * Evacuating is copying (live) objects from a source generation to the target
 * generation.
 *
 * Documentation on the architecture of the Garbage Collector can be
 * found in the online commentary:
 *
 *   https://gitlab.haskell.org/ghc/ghc/wikis/commentary/rts/storage/gc
 *
 * ---------------------------------------------------------------------------*/

#include "rts/PosixSource.h"
#include "Rts.h"

#include "Evac.h"
#include "Storage.h"
#include "GC.h"
#include "GCThread.h"
#include "GCTDecl.h"
#include "GCUtils.h"
#include "Compact.h"
#include "MarkStack.h"
#include "Prelude.h"
#include "Trace.h"
#include "LdvProfile.h"
#include "CNF.h"
#include "Scav.h"
#include "NonMoving.h"
#include "CheckUnload.h" // n_unloaded_objects and markObjectCode

#if defined(THREADED_RTS) && !defined(PARALLEL_GC)
#define evacuate(p) evacuate1(p)
#define evacuate_BLACKHOLE(p) evacuate_BLACKHOLE1(p)
#define HEAP_ALLOCED_GC(p) HEAP_ALLOCED(p)
#endif

#if !defined(PARALLEL_GC) || defined(PROFILING)
#define copy_tag_nolock(p, info, src, size, stp, tag) \
        copy_tag(p, info, src, size, stp, tag)
#endif

/* Note [Selector optimisation depth limit]
 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 * MAX_THUNK_SELECTOR_DEPTH is used to avoid long recursion of
 * eval_thunk_selector due to nested selector thunks. Note that this *only*
 * counts nested selector thunks, e.g. `fst (fst (... (fst x)))`. The collector
 * will traverse interleaved selector-constructor pairs without limit, e.g.
 *
 *     a = (fst b, _)
 *     b = (fst c, _)
 *     c = (fst d, _)
 *     d = (x, _)
 *
 */
#define MAX_THUNK_SELECTOR_DEPTH 16

static void eval_thunk_selector (StgClosure **q, StgSelector *p, bool);
ATTR_NOINLINE static void evacuate_large(StgPtr p);

/* -----------------------------------------------------------------------------
   Allocate some space in which to copy an object.
   -------------------------------------------------------------------------- */

static StgPtr
alloc_in_nonmoving_heap (uint32_t size)
{
    gct->copied += size;
    StgPtr to = nonmovingAllocate(gct->cap, size);

    // See Note [Scavenging the non-moving heap] in NonMovingScav.c.
    // Add segment to the todo list unless it's already there
    // current->todo_link == NULL means not in todo list
    struct NonmovingSegment *seg = nonmovingGetSegment(to);
    if (!seg->todo_link) {
        gen_workspace *ws = &gct->gens[oldest_gen->no];
        seg->todo_link = ws->todo_seg;
        ws->todo_seg = seg;
        bdescr *seg_bd = Bdescr((StgPtr) seg);
        seg_bd->u.scan = to;
    }

    // The object which refers to this closure may have been aged (i.e.
    // retained in a younger generation). Consequently, we must add the
    // closure to the mark queue to ensure that it will be marked.
    //
    // However, if we are in a deadlock detection GC then we disable aging
    // so there is no need.
    //
    // See Note [Non-moving GC: Marking evacuated objects].
    if (major_gc && !deadlock_detect_gc) {
        markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, (StgClosure *) to);
    }
    return to;
}

/* Inlined helper shared between alloc_for_copy_nonmoving and alloc_for_copy. */
STATIC_INLINE StgPtr
alloc_in_moving_heap (uint32_t size, uint32_t gen_no)
{
    gen_workspace *ws = &gct->gens[gen_no];  // zero memory references here

    /* chain a new block onto the to-space for the destination gen if
     * necessary.
     */
    StgPtr to = ws->todo_free;
    ws->todo_free += size;
    if (ws->todo_free > ws->todo_lim) {
        to = todo_block_full(size, ws);
    }
    ASSERT(ws->todo_free >= ws->todo_bd->free && ws->todo_free <= ws->todo_lim);

    return to;
}

/*
 * N.B. We duplicate much of alloc_for_copy here to minimize the number of
 * branches introduced in the moving GC path of alloc_for_copy while minimizing
 * repeated work.
 */
static StgPtr
alloc_for_copy_nonmoving (uint32_t size, uint32_t gen_no)
{
    /* See Note [Deadlock detection under nonmoving collector]. */
    if (deadlock_detect_gc) {
        return alloc_in_nonmoving_heap(size);
    }

    /* Should match logic from alloc_for_copy */
    if (gen_no < gct->evac_gen_no) {
        if (gct->eager_promotion) {
            gen_no = gct->evac_gen_no;
        } else {
            gct->failed_to_evac = true;
        }
    }

    if (gen_no == oldest_gen->no) {
        return alloc_in_nonmoving_heap(size);
    } else {
        return alloc_in_moving_heap(size, gen_no);
    }
}

/* size is in words */
STATIC_INLINE StgPtr
alloc_for_copy (uint32_t size, uint32_t gen_no)
{
    ASSERT(gen_no < RtsFlags.GcFlags.generations);

    if (RTS_UNLIKELY(RtsFlags.GcFlags.useNonmoving)) {
        return alloc_for_copy_nonmoving(size, gen_no);
    }

    /* Find out where we're going, using the handy "to" pointer in
     * the gen of the source object.  If it turns out we need to
     * evacuate to an older generation, adjust it here (see comment
     * by evacuate()).
     */
    if (gen_no < gct->evac_gen_no) {
        if (gct->eager_promotion) {
            gen_no = gct->evac_gen_no;
        } else {
            gct->failed_to_evac = true;
        }
    }

    return alloc_in_moving_heap(size, gen_no);
}

/* -----------------------------------------------------------------------------
   The evacuate() code
   -------------------------------------------------------------------------- */

/*
 * Note [Non-moving GC: Marking evacuated objects]
 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 * When the non-moving collector is in use we must be careful to ensure that any
 * references to objects in the non-moving generation from younger generations
 * are pushed to the mark queue.
 *
 * In particular we need to ensure that we handle newly-promoted objects are
 * correctly marked. For instance, consider this case:
 *
 *     generation 0                          generation 1
 *    ──────────────                        ──────────────
 *
 *                                            ┌───────┐
 *      ┌───────┐                             │   A   │
 *      │   B   │ ◁────────────────────────── │       │
 *      │       │ ──┬─────────────────┐       └───────┘
 *      └───────┘   ┆        after GC │
 *                  ┆                 │
 *      ┌───────┐   ┆ before GC       │       ┌───────┐
 *      │   C   │ ◁┄┘                 └─────▷ │   C'  │
 *      │       │                             │       │
 *      └───────┘                             └───────┘
 *
 *
 * In this case object C started off in generation 0 and was evacuated into
 * generation 1 during the preparatory GC. However, the only reference to C'
 * is from B, which lives in the generation 0 (via aging); this reference will
 * not be visible to the concurrent non-moving collector (which can only
 * traverse the generation 1 heap). Consequently, upon evacuating C we need to
 * ensure that C' is added to the update remembered set as we know that it will
 * continue to be reachable via B (which is assumed to be reachable as it lives
 * in a younger generation).
 *
 * Where this happens depends upon the type of the object (e.g. C'):
 *
 *  - In the case of "normal" small heap-allocated objects this happens in
 *    alloc_for_copy.
 *  - In the case of compact region this happens in evacuate_compact.
 *  - In the case of large objects this happens in evacuate_large.
 *
 * See also Note [Aging under the non-moving collector] in NonMoving.c.
 *
 */

/* size is in words

   We want to *always* inline this as often the size of the closure is static,
   which allows unrolling of the copy loop.

 */
ATTR_ALWAYS_INLINE GNUC_ATTR_HOT static inline void
copy_tag(StgClosure **p, const StgInfoTable *info,
         StgClosure *src, uint32_t size, uint32_t gen_no, StgWord tag)
{
    StgPtr to, from;
    uint32_t i;

    to = alloc_for_copy(size,gen_no);

    from = (StgPtr)src;
    to[0] = (W_)info;
    for (i = 1; i < size; i++) { // unroll for small i
        to[i] = from[i];
    }

//  if (to+size+2 < bd->start + BLOCK_SIZE_W) {
//      __builtin_prefetch(to + size + 2, 1);
//  }

#if defined(PARALLEL_GC)
    {
        const StgInfoTable *new_info;
        new_info = (const StgInfoTable *)cas((StgPtr)&src->header.info, (W_)info, MK_FORWARDING_PTR(to));
        if (new_info != info) {
#if defined(PROFILING)
            // We copied this object at the same time as another
            // thread.  We'll evacuate the object again and the copy
            // we just made will be discarded at the next GC, but we
            // may have copied it after the other thread called
            // SET_EVACUAEE_FOR_LDV(), which would confuse the LDV
            // profiler when it encounters this closure in
            // processHeapClosureForDead.  So we reset the LDVW field
            // here.
            LDVW(to) = 0;
#endif
            return evacuate(p); // does the failed_to_evac stuff
        } else {
            // This doesn't need to have RELEASE ordering since we are guaranteed
            // to scavenge the to-space object on the current core therefore
            // no-one else will follow this pointer (FIXME: Is this true in
            // light of the selector optimization?).
            RELEASE_STORE(p, TAG_CLOSURE(tag,(StgClosure*)to));
        }
    }
#else
    src->header.info = (const StgInfoTable *)MK_FORWARDING_PTR(to);
    *p = TAG_CLOSURE(tag,(StgClosure*)to);
#endif  /* defined(PARALLEL_GC) */

#if defined(PROFILING)
    // We store the size of the just evacuated object in the LDV word so that
    // the profiler can guess the position of the next object later.
    // This is safe only if we are sure that no other thread evacuates
    // the object again, so we cannot use copy_tag_nolock when PROFILING.
    SET_EVACUAEE_FOR_LDV(from, size);
#endif
}

#if defined(PARALLEL_GC) && !defined(PROFILING)
ATTR_ALWAYS_INLINE static inline void
copy_tag_nolock(StgClosure **p, const StgInfoTable *info,
         StgClosure *src, uint32_t size, uint32_t gen_no, StgWord tag)
{
    StgPtr to, from;
    uint32_t i;

    to = alloc_for_copy(size,gen_no);

    from = (StgPtr)src;
    to[0] = (W_)info;
    for (i = 1; i < size; i++) { // unroll for small i
        to[i] = from[i];
    }

    // if somebody else reads the forwarding pointer, we better make
    // sure there's a closure at the end of it.
    RELEASE_STORE(p, TAG_CLOSURE(tag,(StgClosure*)to));
    RELEASE_STORE(&src->header.info, \
                  (const StgInfoTable *)MK_FORWARDING_PTR(to));

//  if (to+size+2 < bd->start + BLOCK_SIZE_W) {
//      __builtin_prefetch(to + size + 2, 1);
//  }

#if defined(PROFILING)
    // We store the size of the just evacuated object in the LDV word so that
    // the profiler can guess the position of the next object later.
    SET_EVACUAEE_FOR_LDV(from, size);
#endif
}
#endif

/* Special version of copy() for when we only want to copy the info
 * pointer of an object, but reserve some padding after it.  This is
 * used to optimise evacuation of TSOs.
 */
ATTR_ALWAYS_INLINE static inline bool
copyPart(StgClosure **p, StgClosure *src, uint32_t size_to_reserve,
         uint32_t size_to_copy, uint32_t gen_no)
{
    StgPtr to, from;
    uint32_t i;
    StgWord info;

#if defined(PARALLEL_GC)
spin:
        info = xchg((StgPtr)&src->header.info, (W_)&stg_WHITEHOLE_info);
        if (info == (W_)&stg_WHITEHOLE_info) {
#if defined(PROF_SPIN)
            whitehole_gc_spin++;
#endif /* PROF_SPIN */
            busy_wait_nop();
            goto spin;
        }
    if (IS_FORWARDING_PTR(info)) {
        RELEASE_STORE(&src->header.info, (const StgInfoTable *)info);
        evacuate(p); // does the failed_to_evac stuff
        return false;
    }
#else
    info = (W_)src->header.info;
#endif /* PARALLEL_GC */

    to = alloc_for_copy(size_to_reserve, gen_no);

    from = (StgPtr)src;
    to[0] = info;
    for (i = 1; i < size_to_copy; i++) { // unroll for small i
        to[i] = from[i];
    }

    RELEASE_STORE(p, (StgClosure *) to);
    RELEASE_STORE(&src->header.info, (const StgInfoTable*)MK_FORWARDING_PTR(to));

#if defined(PROFILING)
    // We store the size of the just evacuated object in the LDV word so that
    // the profiler can guess the position of the next object later.
    SET_EVACUAEE_FOR_LDV(from, size_to_reserve);
    // fill the slop
    if (size_to_reserve - size_to_copy > 0)
        LDV_FILL_SLOP(to + size_to_copy, (int)(size_to_reserve - size_to_copy));
#endif

    return true;
}


/* Copy wrappers that don't tag the closure after copying */
ATTR_ALWAYS_INLINE GNUC_ATTR_HOT static inline void
copy(StgClosure **p, const StgInfoTable *info,
     StgClosure *src, uint32_t size, uint32_t gen_no)
{
    copy_tag(p,info,src,size,gen_no,0);
}

/* -----------------------------------------------------------------------------
   Evacuate a large object

   This just consists of removing the object from the (doubly-linked)
   gen->large_objects list, and linking it on to the (singly-linked)
   gct->todo_large_objects list, from where it will be scavenged later.

   Convention: bd->flags has BF_EVACUATED set for a large object
   that has been evacuated, or unset otherwise.
   -------------------------------------------------------------------------- */

ATTR_NOINLINE static void
evacuate_large(StgPtr p)
{
  bdescr *bd;
  generation *gen, *new_gen;
  uint32_t gen_no, new_gen_no;
  gen_workspace *ws;

  bd = Bdescr(p);
  gen = RELAXED_LOAD(&bd->gen);
  gen_no = RELAXED_LOAD(&bd->gen_no);
  ACQUIRE_SPIN_LOCK(&gen->sync);

  // already evacuated?
  if (RELAXED_LOAD(&bd->flags) & BF_EVACUATED) {
    /* Don't forget to set the gct->failed_to_evac flag if we didn't get
     * the desired destination (see comments in evacuate()).
     */
    if (gen_no < gct->evac_gen_no) {
        gct->failed_to_evac = true;
        TICK_GC_FAILED_PROMOTION();
    }
    RELEASE_SPIN_LOCK(&gen->sync);
    return;
  }

  // remove from large_object list
  dbl_link_remove(bd, &gen->large_objects);

  /* link it on to the evacuated large object list of the destination gen
   */
  new_gen_no = bd->dest_no;

  if (RTS_UNLIKELY(deadlock_detect_gc)) {
      /* See Note [Deadlock detection under nonmoving collector]. */
      new_gen_no = oldest_gen->no;
  } else if (new_gen_no < gct->evac_gen_no) {
      if (gct->eager_promotion) {
          new_gen_no = gct->evac_gen_no;
      } else {
          gct->failed_to_evac = true;
      }
  }

  ws = &gct->gens[new_gen_no];
  new_gen = &generations[new_gen_no];

  __atomic_fetch_or(&bd->flags, BF_EVACUATED, __ATOMIC_ACQ_REL);
  if (RTS_UNLIKELY(RtsFlags.GcFlags.useNonmoving && new_gen == oldest_gen)) {
      __atomic_fetch_or(&bd->flags, BF_NONMOVING, __ATOMIC_ACQ_REL);

      // See Note [Non-moving GC: Marking evacuated objects].
      if (major_gc && !deadlock_detect_gc) {
          markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, (StgClosure *) p);
      }
  }
  initBdescr(bd, new_gen, new_gen->to);

  // If this is a block of pinned or compact objects, we don't have to scan
  // these objects, because they aren't allowed to contain any outgoing
  // pointers.  For these blocks, we skip the scavenge stage and put
  // them straight on the scavenged_large_objects list.
  if (RELAXED_LOAD(&bd->flags) & BF_PINNED) {
      ASSERT(get_itbl((StgClosure *)p)->type == ARR_WORDS);

      if (new_gen != gen) { ACQUIRE_SPIN_LOCK(&new_gen->sync); }
      dbl_link_onto(bd, &new_gen->scavenged_large_objects);
      new_gen->n_scavenged_large_blocks += bd->blocks;
      if (new_gen != gen) { RELEASE_SPIN_LOCK(&new_gen->sync); }
  } else {
      bd->link = ws->todo_large_objects;
      ws->todo_large_objects = bd;
  }

  RELEASE_SPIN_LOCK(&gen->sync);
}

/* ----------------------------------------------------------------------------
   Evacuate static objects

   When a static object is visited for the first time in this GC, it
   is chained on to the gct->static_objects list.

   evacuate_static_object (link_field, q)
     - link_field must be STATIC_LINK(q)
   ------------------------------------------------------------------------- */

STATIC_INLINE void
evacuate_static_object (StgClosure **link_field, StgClosure *q)
{
    if (RTS_UNLIKELY(RtsFlags.GcFlags.useNonmoving)) {
        // See Note [Static objects under the nonmoving collector] in Storage.c.
        if (major_gc && !deadlock_detect_gc)
            markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, q);
        return;
    }

    StgWord link = RELAXED_LOAD((StgWord*) link_field);

    // See Note [STATIC_LINK fields] for how the link field bits work
    if (((link & STATIC_BITS) | prev_static_flag) != 3) {
        StgWord new_list_head = (StgWord)q | static_flag;
#if !defined(THREADED_RTS)
        *link_field = gct->static_objects;
        gct->static_objects = (StgClosure *)new_list_head;
#else
        StgWord prev;
        prev = cas((StgVolatilePtr)link_field, link,
                   (StgWord)gct->static_objects);
        if (prev == link) {
            gct->static_objects = (StgClosure *)new_list_head;
        }
#endif
    }
}

/* ----------------------------------------------------------------------------
   Evacuate an object inside a CompactNFData

   These are treated in a similar way to large objects.  We remove the block
   from the compact_objects list of the generation it is on, and link it onto
   the live_compact_objects list of the destination generation.

   It is assumed that objects in the struct live in the same generation
   as the struct itself all the time.
   ------------------------------------------------------------------------- */
STATIC_INLINE void
evacuate_compact (StgPtr p)
{
    StgCompactNFData *str;
    bdescr *bd;
    generation *gen, *new_gen;
    uint32_t gen_no, new_gen_no;

    // We need to find the Compact# corresponding to this pointer, because it
    // will give us the first block in the compact chain, which is the one we
    // that gets linked onto the compact_objects list.
    str = objectGetCompact((StgClosure*)p);
    ASSERT(get_itbl((StgClosure*)str)->type == COMPACT_NFDATA);

    bd = Bdescr((StgPtr)str);
    gen_no = bd->gen_no;

    if (RELAXED_LOAD(&bd->flags) & BF_NONMOVING) {
        // We may have evacuated the block to the nonmoving generation. If so
        // we need to make sure it is added to the mark queue since the only
        // reference to it may be from the moving heap.
        if (major_gc && !deadlock_detect_gc)
            markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, (StgClosure *) str);
        return;
    }

    // already evacuated? (we're about to do the same check,
    // but we avoid taking the spin-lock)
    if (bd->flags & BF_EVACUATED) {
        /* Don't forget to set the gct->failed_to_evac flag if we didn't get
         * the desired destination (see comments in evacuate()).
         */
        debugTrace(DEBUG_compact, "Compact %p already evacuated", str);
        if (gen_no < gct->evac_gen_no) {
            gct->failed_to_evac = true;
            TICK_GC_FAILED_PROMOTION();
        }
        return;
    }

    gen = bd->gen;
    gen_no = bd->gen_no;
    ACQUIRE_SPIN_LOCK(&gen->sync);

    // already evacuated?
    if (bd->flags & BF_EVACUATED) {
        /* Don't forget to set the gct->failed_to_evac flag if we didn't get
         * the desired destination (see comments in evacuate()).
         */
        if (gen_no < gct->evac_gen_no) {
            gct->failed_to_evac = true;
            TICK_GC_FAILED_PROMOTION();
        }
        RELEASE_SPIN_LOCK(&gen->sync);
        return;
    }

    // remove from compact_objects list
    dbl_link_remove(bd, &gen->compact_objects);

    /* link it on to the evacuated compact object list of the destination gen
     */
    new_gen_no = bd->dest_no;

    if (new_gen_no < gct->evac_gen_no) {
        if (gct->eager_promotion) {
            new_gen_no = gct->evac_gen_no;
        } else {
            gct->failed_to_evac = true;
        }
    }

    new_gen = &generations[new_gen_no];

    // Note: for speed we only update the generation of the first block here
    // This means that bdescr of subsequent blocks will think they are in
    // the wrong generation
    // (This should not be a problem because there is no code that checks
    // for that - the only code touching the generation of the block is
    // in the GC, and that should never see blocks other than the first)
    bd->flags |= BF_EVACUATED;
    if (RTS_UNLIKELY(RtsFlags.GcFlags.useNonmoving && new_gen == oldest_gen)) {
      __atomic_fetch_or(&bd->flags, BF_NONMOVING, __ATOMIC_RELAXED);

      // See Note [Non-moving GC: Marking evacuated objects].
      if (major_gc && !deadlock_detect_gc) {
          markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, (StgClosure *) str);
      }
    }
    initBdescr(bd, new_gen, new_gen->to);

    if (str->hash) {
        // If there is a hash-table for sharing preservation then we need to add
        // the compact to the scavenging work list to ensure that the hashtable
        // is scavenged.
        gen_workspace *ws = &gct->gens[new_gen_no];
        bd->link = ws->todo_large_objects;
        ws->todo_large_objects = bd;
    } else {
        if (new_gen != gen) { ACQUIRE_SPIN_LOCK(&new_gen->sync); }
        dbl_link_onto(bd, &new_gen->live_compact_objects);
        new_gen->n_live_compact_blocks += str->totalW / BLOCK_SIZE_W;
        if (new_gen != gen) { RELEASE_SPIN_LOCK(&new_gen->sync); }
    }

    RELEASE_SPIN_LOCK(&gen->sync);

    // Note: the object did not move in memory, because it lives
    // in pinned (BF_COMPACT) allocation, so we do not need to rewrite it
    // or muck with forwarding pointers
    // Also there is no tag to worry about on the struct (tags are used
    // for constructors and functions, but a struct is neither). There
    // might be a tag on the object pointer, but again we don't change
    // the pointer because we don't move the object so we don't need to
    // rewrite the tag.
}

/* ----------------------------------------------------------------------------
   Evacuate

   Copies an object from its current generation to a target generation. This is
   called (eventually) for every live object in the system.

   The caller to evacuate specifies a desired generation in the
   gct->evac_gen thread-local variable.  The following conditions apply to
   evacuating an object which resides in generation M when we're
   collecting up to generation N

   if  M >= gct->evac_gen
           if  M > N     do nothing
           else          evac to gen->to

   if  M < gct->evac_gen      evac to gct->evac_gen, step 0

   if the object is already evacuated, then we check which generation
   it now resides in.

   if  M >= gct->evac_gen     do nothing
   if  M <  gct->evac_gen     set gct->failed_to_evac flag to indicate that we
                         didn't manage to evacuate this object into gct->evac_gen.


   OPTIMISATION NOTES:

   evacuate() is the single most important function performance-wise
   in the GC.  Various things have been tried to speed it up, but as
   far as I can tell the code generated by gcc 3.2 with -O2 is about
   as good as it's going to get.  We pass the argument to evacuate()
   in a register using the 'regparm' attribute (see the prototype for
   evacuate() near the top of this file).

   Changing evacuate() to take an (StgClosure **) rather than
   returning the new pointer seems attractive, because we can avoid
   writing back the pointer when it hasn't changed (eg. for a static
   object, or an object in a generation > N).  However, I tried it and
   it doesn't help.  One reason is that the (StgClosure **) pointer
   gets spilled to the stack inside evacuate(), resulting in far more
   extra reads/writes than we save.
   ------------------------------------------------------------------------- */

REGPARM1 GNUC_ATTR_HOT void
evacuate(StgClosure **p)
{
  bdescr *bd = NULL;
  uint32_t gen_no;
  StgClosure *q;
  const StgInfoTable *info;
  StgWord tag;

  q = RELAXED_LOAD(p);

loop:
  /* The tag and the pointer are split, to be merged after evacing */
  tag = GET_CLOSURE_TAG(q);
  q = UNTAG_CLOSURE(q);

  ASSERTM(LOOKS_LIKE_CLOSURE_PTR(q), "invalid closure, info=%p", q->header.info);

  if (!HEAP_ALLOCED_GC(q)) {
      if (!major_gc) return;

      // See Note [Object unloading] in CheckUnload.c
      if (RTS_UNLIKELY(unload_mark_needed)) {
          markObjectCode(q);
      }

      info = get_itbl(q);
      switch (info->type) {

      case THUNK_STATIC:
          if (info->srt != 0) {
              evacuate_static_object(THUNK_STATIC_LINK((StgClosure *)q), q);
          }
          return;

      case FUN_STATIC:
          if (info->srt != 0 || info->layout.payload.ptrs != 0) {
              evacuate_static_object(STATIC_LINK(info,(StgClosure *)q), q);
          }
          return;

      case IND_STATIC:
          /* If q->saved_info != NULL, then it's a revertible CAF - it'll be
           * on the CAF list, so don't do anything with it here (we'll
           * scavenge it later).
           */
          evacuate_static_object(IND_STATIC_LINK((StgClosure *)q), q);
          return;

      case CONSTR:
      case CONSTR_1_0:
      case CONSTR_2_0:
      case CONSTR_1_1:
          evacuate_static_object(STATIC_LINK(info,(StgClosure *)q), q);
          return;

      case CONSTR_0_1:
      case CONSTR_0_2:
      case CONSTR_NOCAF:
          /* no need to put these on the static linked list, they don't need
           * to be scavenged.
           */
          return;

      default:
          barf("evacuate(static): strange closure type %d", (int)(info->type));
      }
  }

  bd = Bdescr((P_)q);

  uint16_t flags = RELAXED_LOAD(&bd->flags);
  if ((flags & (BF_LARGE | BF_MARKED | BF_EVACUATED | BF_COMPACT | BF_NONMOVING)) != 0) {
      // Pointer to non-moving heap. Non-moving heap is collected using
      // mark-sweep so this object should be marked and then retained in sweep.
      if (RTS_UNLIKELY(RELAXED_LOAD(&bd->flags) & BF_NONMOVING)) {
          // NOTE: large objects in nonmoving heap are also marked with
          // BF_NONMOVING. Those are moved to scavenged_large_objects list in
          // mark phase.
          if (major_gc && !deadlock_detect_gc)
              markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, q);
          return;
      }

      // pointer into to-space: just return it.  It might be a pointer
      // into a generation that we aren't collecting (> N), or it
      // might just be a pointer into to-space.  The latter doesn't
      // happen often, but allowing it makes certain things a bit
      // easier; e.g. scavenging an object is idempotent, so it's OK to
      // have an object on the mutable list multiple times.
      if (flags & BF_EVACUATED) {
          // We aren't copying this object, so we have to check
          // whether it is already in the target generation.  (this is
          // the write barrier).
          if (RELAXED_LOAD(&bd->gen_no) < gct->evac_gen_no) {
              gct->failed_to_evac = true;
              TICK_GC_FAILED_PROMOTION();
          }
          return;
      }

      // Check for compact before checking for large, this allows doing the
      // right thing for objects that are half way in the middle of the first
      // block of a compact (and would be treated as large objects even though
      // they are not)
      if (flags & BF_COMPACT) {
          evacuate_compact((P_)q);
          return;
      }

      /* evacuate large objects by re-linking them onto a different list.
       */
      if (flags & BF_LARGE) {
          evacuate_large((P_)q);
          return;
      }

      /* If the object is in a gen that we're compacting, then we
       * need to use an alternative evacuate procedure.
       */
      if (!is_marked((P_)q,bd)) {
          mark((P_)q,bd);
          push_mark_stack((P_)q);
      }
      return;
  }

  gen_no = bd->dest_no;

  info = ACQUIRE_LOAD(&q->header.info);
  if (IS_FORWARDING_PTR(info))
  {
    /* Already evacuated, just return the forwarding address.
     * HOWEVER: if the requested destination generation (gct->evac_gen) is
     * older than the actual generation (because the object was
     * already evacuated to a younger generation) then we have to
     * set the gct->failed_to_evac flag to indicate that we couldn't
     * manage to promote the object to the desired generation.
     */
    /*
     * Optimisation: the check is fairly expensive, but we can often
     * shortcut it if either the required generation is 0, or the
     * current object (the EVACUATED) is in a high enough generation.
     * We know that an EVACUATED always points to an object in the
     * same or an older generation.  gen is the lowest generation that the
     * current object would be evacuated to, so we only do the full
     * check if gen is too low.
     */
      StgClosure *e = (StgClosure*)UN_FORWARDING_PTR(info);
      RELAXED_STORE(p, TAG_CLOSURE(tag,e));
      if (gen_no < gct->evac_gen_no) {  // optimisation
          // The ACQUIRE here is necessary to ensure that we see gen_no if the
          // evacuted object lives in a block newly-allocated by a GC thread on
          // another core.
          if (ACQUIRE_LOAD(&Bdescr((P_)e)->gen_no) < gct->evac_gen_no) {
              gct->failed_to_evac = true;
              TICK_GC_FAILED_PROMOTION();
          }
      }
      return;
  }

  switch (INFO_PTR_TO_STRUCT(info)->type) {

  case WHITEHOLE:
      goto loop;

  // For ints and chars of low value, save space by replacing references to
  //    these with closures with references to common, shared ones in the RTS.
  //
  // * Except when compiling into Windows DLLs which don't support cross-package
  //    data references very well.
  //
  case CONSTR_0_1:
  {
#if defined(COMPILING_WINDOWS_DLL)
      copy_tag_nolock(p,info,q,sizeofW(StgHeader)+1,gen_no,tag);
#else
      StgWord w = (StgWord)q->payload[0];
      if (info == Czh_con_info &&
          // unsigned, so always true:  (StgChar)w >= MIN_CHARLIKE &&
          (StgChar)w <= MAX_CHARLIKE) {
          RELAXED_STORE(p, \
                        TAG_CLOSURE(tag, \
                                    (StgClosure *)CHARLIKE_CLOSURE((StgChar)w)
                                   ));
      }
      else if (info == Izh_con_info &&
          (StgInt)w >= MIN_INTLIKE && (StgInt)w <= MAX_INTLIKE) {
          RELAXED_STORE(p, \
                        TAG_CLOSURE(tag, \
                                    (StgClosure *)INTLIKE_CLOSURE((StgInt)w)
                                   ));
      }
      else {
          copy_tag_nolock(p,info,q,sizeofW(StgHeader)+1,gen_no,tag);
      }
#endif
      return;
  }

  case FUN_0_1:
  case FUN_1_0:
  case CONSTR_1_0:
      copy_tag_nolock(p,info,q,sizeofW(StgHeader)+1,gen_no,tag);
      return;

  case THUNK_1_0:
  case THUNK_0_1:
      copy(p,info,q,sizeofW(StgThunk)+1,gen_no);
      return;

  case THUNK_1_1:
  case THUNK_2_0:
  case THUNK_0_2:
    copy(p,info,q,sizeofW(StgThunk)+2,gen_no);
    return;

  case FUN_1_1:
  case FUN_2_0:
  case FUN_0_2:
  case CONSTR_1_1:
  case CONSTR_2_0:
      copy_tag_nolock(p,info,q,sizeofW(StgHeader)+2,gen_no,tag);
      return;

  case CONSTR_0_2:
      copy_tag_nolock(p,info,q,sizeofW(StgHeader)+2,gen_no,tag);
      return;

  case THUNK:
      copy(p,info,q,thunk_sizeW_fromITBL(INFO_PTR_TO_STRUCT(info)),gen_no);
      return;

  case FUN:
  case CONSTR:
  case CONSTR_NOCAF:
      copy_tag_nolock(p,info,q,sizeW_fromITBL(INFO_PTR_TO_STRUCT(info)),gen_no,tag);
      return;

  case BLACKHOLE:
  {
      StgClosure *r;
      const StgInfoTable *i;
      r = ((StgInd*)q)->indirectee;
      if (GET_CLOSURE_TAG(r) == 0) {
          i = ACQUIRE_LOAD(&r->header.info);
          if (IS_FORWARDING_PTR(i)) {
              r = (StgClosure *)UN_FORWARDING_PTR(i);
              i = ACQUIRE_LOAD(&r->header.info);
          }
          if (i == &stg_TSO_info
              || i == &stg_WHITEHOLE_info
              || i == &stg_BLOCKING_QUEUE_CLEAN_info
              || i == &stg_BLOCKING_QUEUE_DIRTY_info) {
              copy(p,info,q,sizeofW(StgInd),gen_no);
              return;
          }
          // Note [BLACKHOLE pointing to IND]
          // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
          // BLOCKING_QUEUE can be overwritten by IND (see
          // wakeBlockingQueue()). However, when this happens we must
          // be updating the BLACKHOLE, so the BLACKHOLE's indirectee
          // should now point to the value.
          //
          // The mutator might observe an inconsistent state, because
          // the writes are happening in another thread, so it's
          // possible for the mutator to follow an indirectee and find
          // an IND. But this should never happen in the GC, because
          // the mutators are all stopped and the writes have
          // completed.
          ASSERT(i != &stg_IND_info);
      }
      q = r;
      RELEASE_STORE(p, r);
      goto loop;
  }

  case MUT_VAR_CLEAN:
  case MUT_VAR_DIRTY:
  case MVAR_CLEAN:
  case MVAR_DIRTY:
  case TVAR:
  case BLOCKING_QUEUE:
  case WEAK:
  case PRIM:
  case MUT_PRIM:
      copy(p,info,q,sizeW_fromITBL(INFO_PTR_TO_STRUCT(info)),gen_no);
      return;

  case BCO:
      copy(p,info,q,bco_sizeW((StgBCO *)q),gen_no);
      return;

  case THUNK_SELECTOR:
      eval_thunk_selector(p, (StgSelector *)q, true);
      return;

  case IND:
    // follow chains of indirections, don't evacuate them
    q = RELAXED_LOAD(&((StgInd*)q)->indirectee);
    RELAXED_STORE(p, q);
    goto loop;

  case RET_BCO:
  case RET_SMALL:
  case RET_BIG:
  case UPDATE_FRAME:
  case UNDERFLOW_FRAME:
  case STOP_FRAME:
  case CATCH_FRAME:
  case CATCH_STM_FRAME:
  case CATCH_RETRY_FRAME:
  case ATOMICALLY_FRAME:
    // shouldn't see these
    barf("evacuate: stack frame at %p\n", q);

  case PAP:
      copy(p,info,q,pap_sizeW((StgPAP*)q),gen_no);
      return;

  case AP:
      copy(p,info,q,ap_sizeW((StgAP*)q),gen_no);
      return;

  case AP_STACK:
      copy(p,info,q,ap_stack_sizeW((StgAP_STACK*)q),gen_no);
      return;

  case ARR_WORDS:
      // just copy the block
      copy(p,info,q,arr_words_sizeW((StgArrBytes *)q),gen_no);
      return;

  case MUT_ARR_PTRS_CLEAN:
  case MUT_ARR_PTRS_DIRTY:
  case MUT_ARR_PTRS_FROZEN_CLEAN:
  case MUT_ARR_PTRS_FROZEN_DIRTY:
      // just copy the block
      copy(p,info,q,mut_arr_ptrs_sizeW((StgMutArrPtrs *)q),gen_no);
      return;

  case SMALL_MUT_ARR_PTRS_CLEAN:
  case SMALL_MUT_ARR_PTRS_DIRTY:
  case SMALL_MUT_ARR_PTRS_FROZEN_CLEAN:
  case SMALL_MUT_ARR_PTRS_FROZEN_DIRTY:
      // just copy the block
      copy(p,info,q,small_mut_arr_ptrs_sizeW((StgSmallMutArrPtrs *)q),gen_no);
      return;

  case TSO:
      copy(p,info,q,sizeofW(StgTSO),gen_no);
      return;

  case STACK:
    {
      StgStack *stack = (StgStack *)q;

      /* To evacuate a small STACK, we need to adjust the stack pointer
       */
      {
          StgStack *new_stack;
          StgPtr r, s;
          bool mine;

          mine = copyPart(p,(StgClosure *)stack, stack_sizeW(stack),
                          sizeofW(StgStack), gen_no);
          if (mine) {
              new_stack = (StgStack *)*p;
              move_STACK(stack, new_stack);
              for (r = stack->sp, s = new_stack->sp;
                   r < stack->stack + stack->stack_size;) {
                  *s++ = *r++;
              }
          }
          return;
      }
    }

  case TREC_CHUNK:
      copy(p,info,q,sizeofW(StgTRecChunk),gen_no);
      return;

  case CONTINUATION:
      copy(p,info,q,continuation_sizeW((StgContinuation*)q),gen_no);
      return;

  default:
    barf("evacuate: strange closure type %d", (int)(INFO_PTR_TO_STRUCT(info)->type));
  }

  barf("evacuate");
}

/* -----------------------------------------------------------------------------
   Evacuate a pointer that is guaranteed to point to a BLACKHOLE.

   This is used for evacuating the updatee of an update frame on the stack.  We
   want to copy the blackhole even if it has been updated by another thread and
   is now an indirection, because the original update frame still needs to
   update it.

   See also Note [upd-black-hole] in sm/Scav.c.
   -------------------------------------------------------------------------- */

void
evacuate_BLACKHOLE(StgClosure **p)
{
    bdescr *bd;
    uint32_t gen_no;
    StgClosure *q;
    const StgInfoTable *info;
    q = *p;

    // closure is required to be a heap-allocated BLACKHOLE
    ASSERT(HEAP_ALLOCED_GC(q));
    ASSERT(GET_CLOSURE_TAG(q) == 0);

    bd = Bdescr((P_)q);
    const uint16_t flags = RELAXED_LOAD(&bd->flags);

    // blackholes can't be in a compact
    ASSERT((flags & BF_COMPACT) == 0);

    if (RTS_UNLIKELY(RELAXED_LOAD(&bd->flags) & BF_NONMOVING)) {
        if (major_gc && !deadlock_detect_gc)
            markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, q);
        return;
    }

    // blackholes *can* be in a large object: when raiseAsync() creates an
    // AP_STACK the payload might be large enough to create a large object.
    // See #14497.
    if (flags & BF_LARGE) {
        evacuate_large((P_)q);
        return;
    }
    if (flags & BF_EVACUATED) {
        if (bd->gen_no < gct->evac_gen_no) {
            gct->failed_to_evac = true;
            TICK_GC_FAILED_PROMOTION();
        }
        return;
    }
    if (flags & BF_MARKED) {
        if (!is_marked((P_)q,bd)) {
            mark((P_)q,bd);
            push_mark_stack((P_)q);
        }
        return;
    }
    gen_no = bd->dest_no;
    info = ACQUIRE_LOAD(&q->header.info);
    if (IS_FORWARDING_PTR(info))
    {
        StgClosure *e = (StgClosure*)UN_FORWARDING_PTR(info);
        *p = e;
        if (gen_no < gct->evac_gen_no) {  // optimisation
            if (ACQUIRE_LOAD(&Bdescr((P_)e)->gen_no) < gct->evac_gen_no) {
                gct->failed_to_evac = true;
                TICK_GC_FAILED_PROMOTION();
            }
        }
        return;
    }

    ASSERT(INFO_PTR_TO_STRUCT(info)->type == BLACKHOLE);
    copy(p,info,q,sizeofW(StgInd),gen_no);
}

/* ----------------------------------------------------------------------------
   Update a chain of thunk selectors with the given value. All selectors in the
   chain become IND pointing to the value, except when there is a loop (i.e.
   the value of a THUNK_SELECTOR is the THUNK_SELECTOR itself), in that case we
   leave the selector as-is.

   p is the current selector to update. In eval_thunk_selector we make a list
   from selectors using ((StgThunk*)p)->payload[0] for the link field and use
   that field to traverse the chain here.

   val is the final value of the selector chain.

   A chain is formed when we've got something like:

      let x = C1 { f1 = e1 }
          y = C2 { f2 = f1 x }
          z = f2 y

   Here the chain (p) we get when evacuating z is:

      [ f2 y, f1 x ]

   and val is e1.
   -------------------------------------------------------------------------- */

static void
unchain_thunk_selectors(StgSelector *p, StgClosure *val)
{
    while (p)
    {
        ASSERT(p->header.info == &stg_WHITEHOLE_info);
        // val must be in to-space.  Not always: when we recursively
        // invoke eval_thunk_selector(), the recursive calls will not
        // evacuate the value (because we want to select on the value,
        // not evacuate it), so in this case val is in from-space.
        // ASSERT(!HEAP_ALLOCED_GC(val) || Bdescr((P_)val)->gen_no > N || (Bdescr((P_)val)->flags & BF_EVACUATED));

        StgSelector *prev = (StgSelector*)((StgClosure *)p)->payload[0];

        // Update the THUNK_SELECTOR with an indirection to the
        // value.  The value is still in from-space at this stage.
        //
        // (old note: Why not do upd_evacuee(q,p)?  Because we have an
        // invariant that an EVACUATED closure always points to an
        // object in the same or an older generation (required by
        // the short-cut test in the EVACUATED case, below).
        if ((StgClosure *)p == val) {
            // must be a loop; just leave a BLACKHOLE in place.  This
            // can happen when we have a chain of selectors that
            // eventually loops back on itself.  We can't leave an
            // indirection pointing to itself, and we want the program
            // to deadlock if it ever enters this closure, so
            // BLACKHOLE is correct.

            // XXX we do not have BLACKHOLEs any more; replace with
            // a THUNK_SELECTOR again.  This will go into a loop if it is
            // entered, and should result in a NonTermination exception.
            RELAXED_STORE(&((StgThunk *)p)->payload[0], val);
            SET_INFO_RELEASE((StgClosure *)p, &stg_sel_0_upd_info);
        } else {
            RELAXED_STORE(&((StgInd *)p)->indirectee, val);
            SET_INFO_RELEASE((StgClosure *)p, &stg_IND_info);
        }

        // For the purposes of LDV profiling, we have created an
        // indirection.
        LDV_RECORD_CREATE(p);

        p = prev;
    }
}

/* -----------------------------------------------------------------------------
   Evaluate a THUNK_SELECTOR if possible.

   p points to a THUNK_SELECTOR that we want to evaluate.

   If the THUNK_SELECTOR could not be evaluated (its selectee is still a THUNK,
   for example), then the THUNK_SELECTOR itself will be evacuated depending on
   the evac parameter.
   -------------------------------------------------------------------------- */

static void
eval_thunk_selector (StgClosure **const q, StgSelector *p, bool evac)
                 // NB. for legacy reasons, p & q are swapped around :(
{
    uint32_t field;
    StgInfoTable *info;
    StgWord info_ptr;
    StgClosure *selectee;
    StgSelector *prev_thunk_selector;
    bdescr *bd;

    prev_thunk_selector = NULL;
    // this is a chain of THUNK_SELECTORs that we are going to update
    // to point to the value of the current THUNK_SELECTOR.  Each
    // closure on the chain is a WHITEHOLE, and points to the next in the
    // chain with payload[0].

selector_chain:

    bd = Bdescr((StgPtr)p);
    if (HEAP_ALLOCED_GC(p)) {
        const uint16_t flags = RELAXED_LOAD(&bd->flags);

        // We should never see large objects here
        ASSERT(!(flags & BF_LARGE));

        // If the THUNK_SELECTOR is in to-space or in a generation that we
        // are not collecting, then bale out early.  We won't be able to
        // save any space in any case, and updating with an indirection is
        // trickier in a non-collected gen: we would have to update the
        // mutable list.
        if (flags & (BF_EVACUATED | BF_NONMOVING)) {
            unchain_thunk_selectors(prev_thunk_selector, (StgClosure *)p);
            if (flags & BF_NONMOVING) {
                // The selector is reachable therefore we must push to the
                // update remembered set in a preparatory collection.
                // See Note [Non-moving GC: Marking evacuated objects].
                markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, (StgClosure*) p);
            }
            *q = (StgClosure *)p;
            // shortcut, behave as for:  if (evac) evacuate(q);
            if (evac && bd->gen_no < gct->evac_gen_no) {
                gct->failed_to_evac = true;
                TICK_GC_FAILED_PROMOTION();
            }
            return;
        }
        // we don't update THUNK_SELECTORS in the compacted
        // generation, because compaction does not remove the INDs
        // that result, this causes confusion later
        // (scavenge_mark_stack doesn't deal with IND).  BEWARE!  This
        // bit is very tricky to get right.  If you make changes
        // around here, test by compiling stage 3 with +RTS -c -RTS.
        if (flags & BF_MARKED) {
            // must call evacuate() to mark this closure if evac==true
            *q = (StgClosure *)p;
            if (evac) evacuate(q);
            unchain_thunk_selectors(prev_thunk_selector, (StgClosure *)p);
            return;
        }
    }


    // WHITEHOLE the selector thunk, since it is now under evaluation.
    // This is important to stop us going into an infinite loop if
    // this selector thunk eventually refers to itself.
#if defined(THREADED_RTS)
    // In threaded mode, we'll use WHITEHOLE to lock the selector
    // thunk while we evaluate it.
    {
        while(true) {
            info_ptr = xchg((StgPtr)&p->header.info, (W_)&stg_WHITEHOLE_info);
            if (info_ptr != (W_)&stg_WHITEHOLE_info) { break; }
#if defined(PROF_SPIN)
            ++whitehole_gc_spin;
#endif
            busy_wait_nop();
        }

        // make sure someone else didn't get here first...
        if (IS_FORWARDING_PTR(info_ptr) ||
            INFO_PTR_TO_STRUCT((StgInfoTable *)info_ptr)->type != THUNK_SELECTOR) {
            // v. tricky now.  The THUNK_SELECTOR has been evacuated
            // by another thread, and is now either a forwarding ptr or IND.
            // We need to extract ourselves from the current situation
            // as cleanly as possible.
            //   - unlock the closure
            //   - update *q, we may have done *some* evaluation
            //   - if evac, we need to call evacuate(), because we
            //     need the write-barrier stuff.
            //   - undo the chain we've built to point to p.
            SET_INFO((StgClosure *)p, (const StgInfoTable *)info_ptr);
            RELEASE_STORE(q, (StgClosure *) p);
            if (Bdescr((StgPtr)p)->flags & BF_NONMOVING) {
                // See Note [Non-moving GC: Marking evacuated objects].
                // TODO: This really shouldn't be necessary since whoever won
                // the race should have pushed
                markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, (StgClosure*) p);
            }
            if (evac) evacuate(q);
            unchain_thunk_selectors(prev_thunk_selector, (StgClosure *)p);
            return;
        }
    }
#else
    // Save the real info pointer (NOTE: not the same as get_itbl()).
    info_ptr = (StgWord)p->header.info;
    SET_INFO((StgClosure *)p,&stg_WHITEHOLE_info);
#endif /* THREADED_RTS */

    field = INFO_PTR_TO_STRUCT((StgInfoTable *)info_ptr)->layout.selector_offset;

    // The selectee might be a constructor closure,
    // so we untag the pointer.
    selectee = UNTAG_CLOSURE(p->selectee);

selector_loop:
    // selectee now points to the closure that we're trying to select
    // a field from.  It may or may not be in to-space: we try not to
    // end up in to-space, but it's impractical to avoid it in
    // general.  The compacting GC scatters to-space pointers in
    // from-space during marking, for example.  We rely on the property
    // that evacuate() doesn't mind if it gets passed a to-space pointer.

    info = RELAXED_LOAD((StgInfoTable**) &selectee->header.info);

    if (IS_FORWARDING_PTR(info)) {
        // We don't follow pointers into to-space; the constructor
        // has already been evacuated, so we won't save any space
        // leaks by evaluating this selector thunk anyhow.
        goto bale_out;
    }

    info = INFO_PTR_TO_STRUCT(info);
    switch (info->type) {
      case WHITEHOLE:
          goto bale_out; // about to be evacuated by another thread (or a loop).

      case CONSTR:
      case CONSTR_1_0:
      case CONSTR_0_1:
      case CONSTR_2_0:
      case CONSTR_1_1:
      case CONSTR_0_2:
      case CONSTR_NOCAF:
          {
              // check that the size is in range
              ASSERT(field <  (StgWord32)(info->layout.payload.ptrs +
                                          info->layout.payload.nptrs));

              // Select the right field from the constructor
              StgClosure *val = RELAXED_LOAD(&selectee->payload[field]);

#if defined(PROFILING)
              // For the purposes of LDV profiling, we have destroyed
              // the original selector thunk, p.
              if (era > 0) {
                  // Only modify the info pointer when LDV profiling is
                  // enabled.  Note that this is incompatible with parallel GC,
                  // because it would allow other threads to start evaluating
                  // the same selector thunk.
                  SET_INFO((StgClosure*)p, (StgInfoTable *)info_ptr);
                  OVERWRITING_CLOSURE((StgClosure*)p);
                  SET_INFO((StgClosure*)p, &stg_WHITEHOLE_info);
                  write_barrier();
#if defined(PARALLEL_GC)
                  abort();  // LDV is incompatible with parallel GC
#endif
              }
#endif

              // the closure in val is now the "value" of the
              // THUNK_SELECTOR in p.  However, val may itself be a
              // THUNK_SELECTOR, in which case we want to continue
              // evaluating until we find the real value, and then
              // update the whole chain to point to the value.
          val_loop:
              info_ptr = ACQUIRE_LOAD((StgWord*) &UNTAG_CLOSURE(val)->header.info);
              if (!IS_FORWARDING_PTR(info_ptr))
              {
                  info = INFO_PTR_TO_STRUCT((StgInfoTable *)info_ptr);
                  switch (info->type) {
                  case IND:
                  case IND_STATIC:
                      val = RELAXED_LOAD(&((StgInd *)val)->indirectee);
                      goto val_loop;
                  case THUNK_SELECTOR:
                      // Use payload to make a list of thunk selectors, to be
                      // used in unchain_thunk_selectors
                      //
                      // FIXME: This seems racy; should we lock this selector to
                      // ensure that another thread doesn't clobber this node
                      // of the chain. This would result in some previous
                      // selectors not being updated when we unchain.
                      RELAXED_STORE(&((StgClosure*)p)->payload[0], (StgClosure *)prev_thunk_selector);
                      prev_thunk_selector = p;
                      p = (StgSelector*)val;
                      goto selector_chain;
                  default:
                      break;
                  }
              }
              RELAXED_STORE(&((StgClosure*)p)->payload[0], (StgClosure *)prev_thunk_selector);
              prev_thunk_selector = p;

              *q = val;

              // update the other selectors in the chain *before*
              // evacuating the value.  This is necessary in the case
              // where the value turns out to be one of the selectors
              // in the chain (i.e. we have a loop), and evacuating it
              // would corrupt the chain.
              unchain_thunk_selectors(prev_thunk_selector, val);

              // evacuate() cannot recurse through
              // eval_thunk_selector(), because we know val is not
              // a THUNK_SELECTOR.
              if (evac) {
                  evacuate(q);
              } else if (isNonmovingClosure(*q)) {
                  // Even if `evac` is false we must ensure that the closure is
                  // pushed since it is reachable.
                  // See Note [Non-moving GC: Marking evacuated objects].
                  markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, (StgClosure*) *q);
              }

              return;
          }

      case IND:
      case IND_STATIC:
          // Again, we might need to untag a constructor.
          selectee = UNTAG_CLOSURE( RELAXED_LOAD(&((StgInd *)selectee)->indirectee) );
          goto selector_loop;

      case BLACKHOLE:
      {
          StgClosure *r;
          const StgInfoTable *i;
          r = ACQUIRE_LOAD(&((StgInd*)selectee)->indirectee);

          // establish whether this BH has been updated, and is now an
          // indirection, as in evacuate().
          if (GET_CLOSURE_TAG(r) == 0) {
              i = ACQUIRE_LOAD(&r->header.info);
              if (IS_FORWARDING_PTR(i)) {
                  r = (StgClosure *)UN_FORWARDING_PTR(i);
                  i = RELAXED_LOAD(&r->header.info);
              }
              if (i == &stg_TSO_info
                  || i == &stg_WHITEHOLE_info
                  || i == &stg_BLOCKING_QUEUE_CLEAN_info
                  || i == &stg_BLOCKING_QUEUE_DIRTY_info) {
                  goto bale_out;
              }
              ASSERT(i != &stg_IND_info);
          }

          selectee = UNTAG_CLOSURE( RELAXED_LOAD(&((StgInd *)selectee)->indirectee) );
          goto selector_loop;
      }

      case THUNK_SELECTOR:
      {
          StgClosure *val;

          // recursively evaluate this selector.  We don't want to
          // recurse indefinitely, so we impose a depth bound.
          // See Note [Selector optimisation depth limit].
          if (gct->thunk_selector_depth >= MAX_THUNK_SELECTOR_DEPTH) {
              if (isNonmovingClosure((StgClosure *) p)) {
                  // See Note [Non-moving GC: Marking evacuated objects].
                  markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, (StgClosure*) p);
              }
              goto bale_out;
          }

          gct->thunk_selector_depth++;
          // false says "don't evacuate the result".  It will,
          // however, update any THUNK_SELECTORs that are evaluated
          // along the way.
          eval_thunk_selector(&val, (StgSelector*)selectee, false);
          gct->thunk_selector_depth--;

          // did we actually manage to evaluate it?
          if (val == selectee) goto bale_out;

          // Of course this pointer might be tagged...
          selectee = UNTAG_CLOSURE(val);
          goto selector_loop;
      }

      case AP:
      case AP_STACK:
      case THUNK:
      case THUNK_1_0:
      case THUNK_0_1:
      case THUNK_2_0:
      case THUNK_1_1:
      case THUNK_0_2:
      case THUNK_STATIC:
          // not evaluated yet
          goto bale_out;

      default:
        barf("eval_thunk_selector: strange selectee %d",
             (int)(info->type));
    }

bale_out:
    // We didn't manage to evaluate this thunk; restore the old info
    // pointer.  But don't forget: we still need to evacuate the thunk itself.
    SET_INFO((StgClosure *)p, (const StgInfoTable *)info_ptr);
    // THREADED_RTS: we just unlocked the thunk, so another thread
    // might get in and update it.  copy() will lock it again and
    // check whether it was updated in the meantime.
    *q = (StgClosure *)p;
    if (evac) {
        copy(q,(const StgInfoTable *)info_ptr,(StgClosure *)p,THUNK_SELECTOR_sizeW(),bd->dest_no);
    }
    if (isNonmovingClosure(*q)) {
        // See Note [Non-moving GC: Marking evacuated objects].
        markQueuePushClosureGC(&gct->cap->upd_rem_set.queue, *q);
    }
    unchain_thunk_selectors(prev_thunk_selector, *q);
}