RetainerProfile.c 65.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
/* -----------------------------------------------------------------------------
 *
 * (c) The GHC Team, 2001
 * Author: Sungwoo Park
 *
 * Retainer profiling.
 *
 * ---------------------------------------------------------------------------*/

#ifdef PROFILING

12 13 14 15 16 17 18
// Turn off inlining when debugging - it obfuscates things
#ifdef DEBUG
#define INLINE
#else
#define INLINE inline
#endif

19 20 21 22 23 24 25 26 27 28 29 30 31 32
#include "Rts.h"
#include "RtsUtils.h"
#include "RetainerProfile.h"
#include "RetainerSet.h"
#include "Schedule.h"
#include "Printer.h"
#include "Storage.h"
#include "RtsFlags.h"
#include "Weak.h"
#include "Sanity.h"
#include "Profiling.h"
#include "Stats.h"
#include "BlockAlloc.h"
#include "ProfHeap.h"
33
#include "Apply.h"
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

/*
  Note: what to change in order to plug-in a new retainer profiling scheme?
    (1) type retainer in ../includes/StgRetainerProf.h
    (2) retainer function R(), i.e., getRetainerFrom()
    (3) the two hashing functions, hashKeySingleton() and hashKeyAddElement(),
        in RetainerSet.h, if needed.
    (4) printRetainer() and printRetainerSetShort() in RetainerSet.c.
 */

/* -----------------------------------------------------------------------------
 * Declarations...
 * -------------------------------------------------------------------------- */

static nat retainerGeneration;	// generation

static nat numObjectVisited;	// total number of objects visited
static nat timesAnyObjectVisited; // number of times any objects are visited

/*
  The rs field in the profile header of any object points to its retainer
  set in an indirect way: if flip is 0, it points to the retainer set;
  if flip is 1, it points to the next byte after the retainer set (even
  for NULL pointers). Therefore, with flip 1, (rs ^ 1) is the actual
  pointer. See retainerSetOf().
 */

61
StgWord flip = 0;     // flip bit
62 63 64 65 66
                      // must be 0 if DEBUG_RETAINER is on (for static closures)

#define setRetainerSetToNull(c)   \
  (c)->header.prof.hp.rs = (RetainerSet *)((StgWord)NULL | flip)

67
static void retainStack(StgClosure *, retainer, StgPtr, StgPtr);
68
static void retainClosure(StgClosure *, StgClosure *, retainer);
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108
#ifdef DEBUG_RETAINER
static void belongToHeap(StgPtr p);
#endif

#ifdef DEBUG_RETAINER
/*
  cStackSize records how many times retainStack() has been invoked recursively,
  that is, the number of activation records for retainStack() on the C stack.
  maxCStackSize records its max value.
  Invariants:
    cStackSize <= maxCStackSize
 */
static nat cStackSize, maxCStackSize;

static nat sumOfNewCost;	// sum of the cost of each object, computed
				// when the object is first visited
static nat sumOfNewCostExtra;   // for those objects not visited during
                                // retainer profiling, e.g., MUT_VAR
static nat costArray[N_CLOSURE_TYPES];

nat sumOfCostLinear;		// sum of the costs of all object, computed
				// when linearly traversing the heap after
				// retainer profiling
nat costArrayLinear[N_CLOSURE_TYPES];
#endif

/* -----------------------------------------------------------------------------
 * Retainer stack - header
 *   Note:
 *     Although the retainer stack implementation could be separated *
 *     from the retainer profiling engine, there does not seem to be
 *     any advantage in doing that; retainer stack is an integral part
 *     of retainer profiling engine and cannot be use elsewhere at
 *     all.
 * -------------------------------------------------------------------------- */

typedef enum {
    posTypeStep,
    posTypePtrs,
    posTypeSRT,
109
    posTypeLargeSRT,
110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131
} nextPosType;

typedef union {
    // fixed layout or layout specified by a field in the closure
    StgWord step;

    // layout.payload
    struct {
    // See StgClosureInfo in InfoTables.h
#if SIZEOF_VOID_P == 8
	StgWord32 pos;
	StgWord32 ptrs;
#else
	StgWord16 pos;
	StgWord16 ptrs;
#endif
	StgPtr payload;
    } ptrs;

    // SRT
    struct {
	StgClosure **srt;
132
	StgWord    srt_bitmap;
133
    } srt;
134 135 136 137 138 139 140

    // Large SRT
    struct {
	StgLargeSRT *srt;
	StgWord offset;
    } large_srt;
	
141 142 143 144 145 146 147 148 149
} nextPos;

typedef struct {
    nextPosType type;
    nextPos next;
} stackPos;

typedef struct {
    StgClosure *c;
150
    retainer c_child_r;
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
    stackPos info;
} stackElement;

/*
  Invariants:
    firstStack points to the first block group.
    currentStack points to the block group currently being used.
    currentStack->free == stackLimit.
    stackTop points to the topmost byte in the stack of currentStack.
    Unless the whole stack is empty, stackTop must point to the topmost
    object (or byte) in the whole stack. Thus, it is only when the whole stack
    is empty that stackTop == stackLimit (not during the execution of push()
    and pop()).
    stackBottom == currentStack->start.
    stackLimit == currentStack->start + BLOCK_SIZE_W * currentStack->blocks.
  Note:
    When a current stack becomes empty, stackTop is set to point to
    the topmost element on the previous block group so as to satisfy
    the invariants described above.
 */
sof's avatar
sof committed
171
static bdescr *firstStack = NULL;
172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
static bdescr *currentStack;
static stackElement *stackBottom, *stackTop, *stackLimit;

/*
  currentStackBoundary is used to mark the current stack chunk.
  If stackTop == currentStackBoundary, it means that the current stack chunk
  is empty. It is the responsibility of the user to keep currentStackBoundary
  valid all the time if it is to be employed.
 */
static stackElement *currentStackBoundary;

/*
  stackSize records the current size of the stack.
  maxStackSize records its high water mark.
  Invariants:
    stackSize <= maxStackSize
  Note:
    stackSize is just an estimate measure of the depth of the graph. The reason
    is that some heap objects have only a single child and may not result
    in a new element being pushed onto the stack. Therefore, at the end of
    retainer profiling, maxStackSize + maxCStackSize is some value no greater
    than the actual depth of the graph.
 */
#ifdef DEBUG_RETAINER
static int stackSize, maxStackSize;
#endif

// number of blocks allocated for one stack
#define BLOCKS_IN_STACK 1

/* -----------------------------------------------------------------------------
 * Add a new block group to the stack.
 * Invariants:
 *  currentStack->link == s.
 * -------------------------------------------------------------------------- */
207
static INLINE void
208 209 210 211 212 213 214 215 216 217 218 219 220 221
newStackBlock( bdescr *bd )
{
    currentStack = bd;
    stackTop     = (stackElement *)(bd->start + BLOCK_SIZE_W * bd->blocks);
    stackBottom  = (stackElement *)bd->start;
    stackLimit   = (stackElement *)stackTop;
    bd->free     = (StgPtr)stackLimit;
}

/* -----------------------------------------------------------------------------
 * Return to the previous block group.
 * Invariants:
 *   s->link == currentStack.
 * -------------------------------------------------------------------------- */
222
static INLINE void
223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
returnToOldStack( bdescr *bd )
{
    currentStack = bd;
    stackTop = (stackElement *)bd->free;
    stackBottom = (stackElement *)bd->start;
    stackLimit = (stackElement *)(bd->start + BLOCK_SIZE_W * bd->blocks);
    bd->free = (StgPtr)stackLimit;
}

/* -----------------------------------------------------------------------------
 *  Initializes the traverse stack.
 * -------------------------------------------------------------------------- */
static void
initializeTraverseStack( void )
{
    if (firstStack != NULL) {
	freeChain(firstStack);
    }

    firstStack = allocGroup(BLOCKS_IN_STACK);
    firstStack->link = NULL;
    firstStack->u.back = NULL;

    newStackBlock(firstStack);
}

/* -----------------------------------------------------------------------------
 * Frees all the block groups in the traverse stack.
 * Invariants:
 *   firstStack != NULL
 * -------------------------------------------------------------------------- */
static void
closeTraverseStack( void )
{
    freeChain(firstStack);
    firstStack = NULL;
}

/* -----------------------------------------------------------------------------
 * Returns rtsTrue if the whole stack is empty.
 * -------------------------------------------------------------------------- */
264
static INLINE rtsBool
265 266 267 268 269
isEmptyRetainerStack( void )
{
    return (firstStack == currentStack) && stackTop == stackLimit;
}

sof's avatar
sof committed
270 271 272
/* -----------------------------------------------------------------------------
 * Returns size of stack
 * -------------------------------------------------------------------------- */
273
#ifdef DEBUG
sof's avatar
sof committed
274
lnat
275
retainerStackBlocks( void )
sof's avatar
sof committed
276 277 278 279 280 281 282 283 284
{
    bdescr* bd;
    lnat res = 0;

    for (bd = firstStack; bd != NULL; bd = bd->link) 
      res += bd->blocks;

    return res;
}
285
#endif
sof's avatar
sof committed
286

287 288 289 290
/* -----------------------------------------------------------------------------
 * Returns rtsTrue if stackTop is at the stack boundary of the current stack,
 * i.e., if the current stack chunk is empty.
 * -------------------------------------------------------------------------- */
291
static INLINE rtsBool
292 293 294 295 296 297 298 299 300 301
isOnBoundary( void )
{
    return stackTop == currentStackBoundary;
}

/* -----------------------------------------------------------------------------
 * Initializes *info from ptrs and payload.
 * Invariants:
 *   payload[] begins with ptrs pointers followed by non-pointers.
 * -------------------------------------------------------------------------- */
302
static INLINE void
303 304 305 306 307 308 309 310 311 312 313
init_ptrs( stackPos *info, nat ptrs, StgPtr payload )
{
    info->type              = posTypePtrs;
    info->next.ptrs.pos     = 0;
    info->next.ptrs.ptrs    = ptrs;
    info->next.ptrs.payload = payload;
}

/* -----------------------------------------------------------------------------
 * Find the next object from *info.
 * -------------------------------------------------------------------------- */
314
static INLINE StgClosure *
315 316 317 318 319 320 321 322 323 324 325 326
find_ptrs( stackPos *info )
{
    if (info->next.ptrs.pos < info->next.ptrs.ptrs) {
	return (StgClosure *)info->next.ptrs.payload[info->next.ptrs.pos++];
    } else {
	return NULL;
    }
}

/* -----------------------------------------------------------------------------
 *  Initializes *info from SRT information stored in *infoTable.
 * -------------------------------------------------------------------------- */
327
static INLINE void
328
init_srt_fun( stackPos *info, StgFunInfoTable *infoTable )
329
{
330 331
    if (infoTable->i.srt_bitmap == (StgHalfWord)(-1)) {
	info->type = posTypeLargeSRT;
332
	info->next.large_srt.srt = (StgLargeSRT *)GET_FUN_SRT(infoTable);
333 334 335
	info->next.large_srt.offset = 0;
    } else {
	info->type = posTypeSRT;
336
	info->next.srt.srt = (StgClosure **)GET_FUN_SRT(infoTable);
337 338
	info->next.srt.srt_bitmap = infoTable->i.srt_bitmap;
    }
339 340
}

341
static INLINE void
342 343
init_srt_thunk( stackPos *info, StgThunkInfoTable *infoTable )
{
344 345
    if (infoTable->i.srt_bitmap == (StgHalfWord)(-1)) {
	info->type = posTypeLargeSRT;
346
	info->next.large_srt.srt = (StgLargeSRT *)GET_SRT(infoTable);
347 348 349
	info->next.large_srt.offset = 0;
    } else {
	info->type = posTypeSRT;
350
	info->next.srt.srt = (StgClosure **)GET_SRT(infoTable);
351 352
	info->next.srt.srt_bitmap = infoTable->i.srt_bitmap;
    }
353 354 355 356 357
}

/* -----------------------------------------------------------------------------
 * Find the next object from *info.
 * -------------------------------------------------------------------------- */
358
static INLINE StgClosure *
359 360 361
find_srt( stackPos *info )
{
    StgClosure *c;
362
    StgWord bitmap;
363

364 365 366 367 368
    if (info->type == posTypeSRT) {
	// Small SRT bitmap
	bitmap = info->next.srt.srt_bitmap;
	while (bitmap != 0) {
	    if ((bitmap & 1) != 0) {
369
#ifdef ENABLE_WIN32_DLL_SUPPORT
370 371 372 373 374
		
		if ((unsigned long)(*(info->next.srt.srt)) & 0x1)
		    c = (* (StgClosure **)((unsigned long)*(info->next.srt.srt)) & ~0x1);
		else
		    c = *(info->next.srt.srt);
375
#else
376
		c = *(info->next.srt.srt);
377
#endif
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412
		bitmap = bitmap >> 1;
		info->next.srt.srt++;
		info->next.srt.srt_bitmap = bitmap;
		return c;
	    }
	    bitmap = bitmap >> 1;
	    info->next.srt.srt++;
	}
	// bitmap is now zero...
	return NULL;
    }
    else {
	// Large SRT bitmap
	nat i = info->next.large_srt.offset;
	StgWord bitmap;

	// Follow the pattern from GC.c:scavenge_large_srt_bitmap().
	bitmap = info->next.large_srt.srt->l.bitmap[i / BITS_IN(W_)];
	bitmap = bitmap >> (i % BITS_IN(StgWord));
	while (i < info->next.large_srt.srt->l.size) {
	    if ((bitmap & 1) != 0) {
		c = ((StgClosure **)info->next.large_srt.srt->srt)[i];
		i++;
		info->next.large_srt.offset = i;
		return c;
	    }
	    i++;
	    if (i % BITS_IN(W_) == 0) {
		bitmap = info->next.large_srt.srt->l.bitmap[i / BITS_IN(W_)];
	    } else {
		bitmap = bitmap >> 1;
	    }
	}
	// reached the end of this bitmap.
	info->next.large_srt.offset = i;
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427
	return NULL;
    }
}

/* -----------------------------------------------------------------------------
 *  push() pushes a stackElement representing the next child of *c
 *  onto the traverse stack. If *c has no child, *first_child is set
 *  to NULL and nothing is pushed onto the stack. If *c has only one
 *  child, *c_chlid is set to that child and nothing is pushed onto
 *  the stack.  If *c has more than two children, *first_child is set
 *  to the first child and a stackElement representing the second
 *  child is pushed onto the stack.

 *  Invariants:
 *     *c_child_r is the most recent retainer of *c's children.
428
 *     *c is not any of TSO, AP, PAP, AP_STACK, which means that
429 430 431
 *        there cannot be any stack objects.
 *  Note: SRTs are considered to  be children as well.
 * -------------------------------------------------------------------------- */
432
static INLINE void
433
push( StgClosure *c, retainer c_child_r, StgClosure **first_child )
434 435 436 437 438
{
    stackElement se;
    bdescr *nbd;      // Next Block Descriptor

#ifdef DEBUG_RETAINER
439
    // debugBelch("push(): stackTop = 0x%x, currentStackBoundary = 0x%x\n", stackTop, currentStackBoundary);
440 441 442
#endif

    ASSERT(get_itbl(c)->type != TSO);
443
    ASSERT(get_itbl(c)->type != AP_STACK);
444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465

    //
    // fill in se
    //

    se.c = c;
    se.c_child_r = c_child_r;

    // fill in se.info
    switch (get_itbl(c)->type) {
	// no child, no SRT
    case CONSTR_0_1:
    case CONSTR_0_2:
    case CAF_BLACKHOLE:
    case BLACKHOLE:
    case SE_BLACKHOLE:
    case SE_CAF_BLACKHOLE:
    case ARR_WORDS:
	*first_child = NULL;
	return;

	// one child (fixed), no SRT
466 467
    case MUT_VAR_CLEAN:
    case MUT_VAR_DIRTY:
468 469 470 471 472 473 474 475
	*first_child = ((StgMutVar *)c)->var;
	return;
    case THUNK_SELECTOR:
	*first_child = ((StgSelector *)c)->selectee;
	return;
    case IND_PERM:
    case IND_OLDGEN_PERM:
    case IND_OLDGEN:
476
	*first_child = ((StgInd *)c)->indirectee;
477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524
	return;
    case CONSTR_1_0:
    case CONSTR_1_1:
	*first_child = c->payload[0];
	return;

	// For CONSTR_2_0 and MVAR, we use se.info.step to record the position
	// of the next child. We do not write a separate initialization code.
	// Also we do not have to initialize info.type;

	// two children (fixed), no SRT
	// need to push a stackElement, but nothing to store in se.info
    case CONSTR_2_0:
	*first_child = c->payload[0];         // return the first pointer
	// se.info.type = posTypeStep;
	// se.info.next.step = 2;            // 2 = second
	break;

	// three children (fixed), no SRT
	// need to push a stackElement
    case MVAR:
	// head must be TSO and the head of a linked list of TSOs.
	// Shoule it be a child? Seems to be yes.
	*first_child = (StgClosure *)((StgMVar *)c)->head;
	// se.info.type = posTypeStep;
	se.info.next.step = 2;            // 2 = second
	break;

	// three children (fixed), no SRT
    case WEAK:
	*first_child = ((StgWeak *)c)->key;
	// se.info.type = posTypeStep;
	se.info.next.step = 2;
	break;

	// layout.payload.ptrs, no SRT
    case CONSTR:
    case STABLE_NAME:
    case BCO:
    case CONSTR_STATIC:
	init_ptrs(&se.info, get_itbl(c)->layout.payload.ptrs,
		  (StgPtr)c->payload);
	*first_child = find_ptrs(&se.info);
	if (*first_child == NULL)
	    return;   // no child
	break;

	// StgMutArrPtr.ptrs, no SRT
525 526
    case MUT_ARR_PTRS_CLEAN:
    case MUT_ARR_PTRS_DIRTY:
527
    case MUT_ARR_PTRS_FROZEN:
528
    case MUT_ARR_PTRS_FROZEN0:
529 530 531 532 533 534 535 536 537 538
	init_ptrs(&se.info, ((StgMutArrPtrs *)c)->ptrs,
		  (StgPtr)(((StgMutArrPtrs *)c)->payload));
	*first_child = find_ptrs(&se.info);
	if (*first_child == NULL)
	    return;
	break;

    // layout.payload.ptrs, SRT
    case FUN:           // *c is a heap object.
    case FUN_2_0:
539 540 541 542 543 544 545
	init_ptrs(&se.info, get_itbl(c)->layout.payload.ptrs, (StgPtr)c->payload);
	*first_child = find_ptrs(&se.info);
	if (*first_child == NULL)
	    // no child from ptrs, so check SRT
	    goto fun_srt_only;
	break;

546 547
    case THUNK:
    case THUNK_2_0:
548 549
	init_ptrs(&se.info, get_itbl(c)->layout.payload.ptrs, 
		  (StgPtr)((StgThunk *)c)->payload);
550 551 552
	*first_child = find_ptrs(&se.info);
	if (*first_child == NULL)
	    // no child from ptrs, so check SRT
553
	    goto thunk_srt_only;
554 555 556 557 558
	break;

	// 1 fixed child, SRT
    case FUN_1_0:
    case FUN_1_1:
559 560 561 562 563
	*first_child = c->payload[0];
	ASSERT(*first_child != NULL);
	init_srt_fun(&se.info, get_fun_itbl(c));
	break;

564 565
    case THUNK_1_0:
    case THUNK_1_1:
566
	*first_child = ((StgThunk *)c)->payload[0];
567
	ASSERT(*first_child != NULL);
568
	init_srt_thunk(&se.info, get_thunk_itbl(c));
569 570 571
	break;

    case FUN_STATIC:      // *c is a heap object.
572
	ASSERT(get_itbl(c)->srt_bitmap != 0);
573 574
    case FUN_0_1:
    case FUN_0_2:
575 576 577 578 579 580 581 582 583
    fun_srt_only:
        init_srt_fun(&se.info, get_fun_itbl(c));
	*first_child = find_srt(&se.info);
	if (*first_child == NULL)
	    return;     // no child
	break;

    // SRT only
    case THUNK_STATIC:
584
	ASSERT(get_itbl(c)->srt_bitmap != 0);
585 586
    case THUNK_0_1:
    case THUNK_0_2:
587 588
    thunk_srt_only:
        init_srt_thunk(&se.info, get_thunk_itbl(c));
589 590 591 592
	*first_child = find_srt(&se.info);
	if (*first_child == NULL)
	    return;     // no child
	break;
593 594 595 596 597 598 599 600 601 602 603 604 605 606 607
	
    case TVAR_WAIT_QUEUE:
	*first_child = (StgClosure *)((StgTVarWaitQueue *)c)->waiting_tso;
	se.info.next.step = 2;            // 2 = second
	break;
    case TVAR:
	*first_child = (StgClosure *)((StgTVar *)c)->current_value;
	break;
    case TREC_HEADER:
	*first_child = (StgClosure *)((StgTRecHeader *)c)->enclosing_trec;
	break;
    case TREC_CHUNK:
	*first_child = (StgClosure *)((StgTRecChunk *)c)->prev_chunk;
	se.info.next.step = 0;  // entry no.
	break;
608 609 610

	// cannot appear
    case PAP:
611 612
    case AP:
    case AP_STACK:
613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643
    case TSO:
    case IND_STATIC:
    case CONSTR_INTLIKE:
    case CONSTR_CHARLIKE:
    case CONSTR_NOCAF_STATIC:
	// stack objects
    case UPDATE_FRAME:
    case CATCH_FRAME:
    case STOP_FRAME:
    case RET_DYN:
    case RET_BCO:
    case RET_SMALL:
    case RET_VEC_SMALL:
    case RET_BIG:
    case RET_VEC_BIG:
	// invalid objects
    case IND:
    case BLOCKED_FETCH:
    case FETCH_ME:
    case FETCH_ME_BQ:
    case RBH:
    case REMOTE_REF:
    case EVACUATED:
    case INVALID_OBJECT:
    default:
	barf("Invalid object *c in push()");
	return;
    }

    if (stackTop - 1 < stackBottom) {
#ifdef DEBUG_RETAINER
644
	// debugBelch("push() to the next stack.\n");
645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672
#endif
	// currentStack->free is updated when the active stack is switched
	// to the next stack.
	currentStack->free = (StgPtr)stackTop;

	if (currentStack->link == NULL) {
	    nbd = allocGroup(BLOCKS_IN_STACK);
	    nbd->link = NULL;
	    nbd->u.back = currentStack;
	    currentStack->link = nbd;
	} else
	    nbd = currentStack->link;

	newStackBlock(nbd);
    }

    // adjust stackTop (acutal push)
    stackTop--;
    // If the size of stackElement was huge, we would better replace the
    // following statement by either a memcpy() call or a switch statement
    // on the type of the element. Currently, the size of stackElement is
    // small enough (5 words) that this direct assignment seems to be enough.
    *stackTop = se;

#ifdef DEBUG_RETAINER
    stackSize++;
    if (stackSize > maxStackSize) maxStackSize = stackSize;
    // ASSERT(stackSize >= 0);
673
    // debugBelch("stackSize = %d\n", stackSize);
674 675 676 677 678 679 680 681 682 683 684 685 686
#endif
}

/* -----------------------------------------------------------------------------
 *  popOff() and popOffReal(): Pop a stackElement off the traverse stack.
 *  Invariants:
 *    stackTop cannot be equal to stackLimit unless the whole stack is
 *    empty, in which case popOff() is not allowed.
 *  Note:
 *    You can think of popOffReal() as a part of popOff() which is
 *    executed at the end of popOff() in necessary. Since popOff() is
 *    likely to be executed quite often while popOffReal() is not, we
 *    separate popOffReal() from popOff(), which is declared as an
687
 *    INLINE function (for the sake of execution speed).  popOffReal()
688 689 690 691 692 693 694 695
 *    is called only within popOff() and nowhere else.
 * -------------------------------------------------------------------------- */
static void
popOffReal(void)
{
    bdescr *pbd;    // Previous Block Descriptor

#ifdef DEBUG_RETAINER
696
    // debugBelch("pop() to the previous stack.\n");
697 698 699 700 701 702 703 704 705 706 707 708 709 710
#endif

    ASSERT(stackTop + 1 == stackLimit);
    ASSERT(stackBottom == (stackElement *)currentStack->start);

    if (firstStack == currentStack) {
	// The stack is completely empty.
	stackTop++;
	ASSERT(stackTop == stackLimit);
#ifdef DEBUG_RETAINER
	stackSize--;
	if (stackSize > maxStackSize) maxStackSize = stackSize;
	/*
	  ASSERT(stackSize >= 0);
711
	  debugBelch("stackSize = %d\n", stackSize);
712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731
	*/
#endif
	return;
    }

    // currentStack->free is updated when the active stack is switched back
    // to the previous stack.
    currentStack->free = (StgPtr)stackLimit;

    // find the previous block descriptor
    pbd = currentStack->u.back;
    ASSERT(pbd != NULL);

    returnToOldStack(pbd);

#ifdef DEBUG_RETAINER
    stackSize--;
    if (stackSize > maxStackSize) maxStackSize = stackSize;
    /*
      ASSERT(stackSize >= 0);
732
      debugBelch("stackSize = %d\n", stackSize);
733 734 735 736
    */
#endif
}

737
static INLINE void
738 739
popOff(void) {
#ifdef DEBUG_RETAINER
740
    // debugBelch("\tpopOff(): stackTop = 0x%x, currentStackBoundary = 0x%x\n", stackTop, currentStackBoundary);
741 742 743 744 745 746 747 748 749 750 751 752 753
#endif

    ASSERT(stackTop != stackLimit);
    ASSERT(!isEmptyRetainerStack());

    // <= (instead of <) is wrong!
    if (stackTop + 1 < stackLimit) {
	stackTop++;
#ifdef DEBUG_RETAINER
	stackSize--;
	if (stackSize > maxStackSize) maxStackSize = stackSize;
	/*
	  ASSERT(stackSize >= 0);
754
	  debugBelch("stackSize = %d\n", stackSize);
755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778
	*/
#endif
	return;
    }

    popOffReal();
}

/* -----------------------------------------------------------------------------
 *  Finds the next object to be considered for retainer profiling and store
 *  its pointer to *c.
 *  Test if the topmost stack element indicates that more objects are left,
 *  and if so, retrieve the first object and store its pointer to *c. Also,
 *  set *cp and *r appropriately, both of which are stored in the stack element.
 *  The topmost stack element then is overwritten so as for it to now denote
 *  the next object.
 *  If the topmost stack element indicates no more objects are left, pop
 *  off the stack element until either an object can be retrieved or
 *  the current stack chunk becomes empty, indicated by rtsTrue returned by
 *  isOnBoundary(), in which case *c is set to NULL.
 *  Note:
 *    It is okay to call this function even when the current stack chunk
 *    is empty.
 * -------------------------------------------------------------------------- */
779
static INLINE void
780
pop( StgClosure **c, StgClosure **cp, retainer *r )
781 782 783 784
{
    stackElement *se;

#ifdef DEBUG_RETAINER
785
    // debugBelch("pop(): stackTop = 0x%x, currentStackBoundary = 0x%x\n", stackTop, currentStackBoundary);
786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834
#endif

    do {
	if (isOnBoundary()) {     // if the current stack chunk is depleted
	    *c = NULL;
	    return;
	}

	se = stackTop;

	switch (get_itbl(se->c)->type) {
	    // two children (fixed), no SRT
	    // nothing in se.info
	case CONSTR_2_0:
	    *c = se->c->payload[1];
	    *cp = se->c;
	    *r = se->c_child_r;
	    popOff();
	    return;

	    // three children (fixed), no SRT
	    // need to push a stackElement
	case MVAR:
	    if (se->info.next.step == 2) {
		*c = (StgClosure *)((StgMVar *)se->c)->tail;
		se->info.next.step++;             // move to the next step
		// no popOff
	    } else {
		*c = ((StgMVar *)se->c)->value;
		popOff();
	    }
	    *cp = se->c;
	    *r = se->c_child_r;
	    return;

	    // three children (fixed), no SRT
	case WEAK:
	    if (se->info.next.step == 2) {
		*c = ((StgWeak *)se->c)->value;
		se->info.next.step++;
		// no popOff
	    } else {
		*c = ((StgWeak *)se->c)->finalizer;
		popOff();
	    }
	    *cp = se->c;
	    *r = se->c_child_r;
	    return;

835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888
	case TVAR_WAIT_QUEUE:
	    if (se->info.next.step == 2) {
		*c = (StgClosure *)((StgTVarWaitQueue *)se->c)->next_queue_entry;
		se->info.next.step++;             // move to the next step
		// no popOff
	    } else {
		*c = (StgClosure *)((StgTVarWaitQueue *)se->c)->prev_queue_entry;
		popOff();
	    }
	    *cp = se->c;
	    *r = se->c_child_r;
	    return;

	case TVAR:
	    *c = (StgClosure *)((StgTVar *)se->c)->first_wait_queue_entry;
	    *cp = se->c;
	    *r = se->c_child_r;
	    popOff();
	    return;

	case TREC_HEADER:
	    *c = (StgClosure *)((StgTRecHeader *)se->c)->current_chunk;
	    *cp = se->c;
	    *r = se->c_child_r;
	    popOff();
	    return;

	case TREC_CHUNK: {
	    // These are pretty complicated: we have N entries, each
	    // of which contains 3 fields that we want to follow.  So
	    // we divide the step counter: the 2 low bits indicate
	    // which field, and the rest of the bits indicate the
	    // entry number (starting from zero).
	    nat entry_no = se->info.next.step >> 2;
	    nat field_no = se->info.next.step & 3;
	    if (entry_no == ((StgTRecChunk *)se->c)->next_entry_idx) {
		*c = NULL;
		popOff();
		return;
	    }
	    TRecEntry *entry = &((StgTRecChunk *)se->c)->entries[entry_no];
	    if (field_no == 0) {
		*c = (StgClosure *)entry->tvar;
	    } else if (field_no == 1) {
		*c = entry->expected_value;
	    } else {
		*c = entry->new_value;
	    }
	    *cp = se->c;
	    *r = se->c_child_r;
	    se->info.next.step++;
	    return;
	}

889 890 891 892 893
	case CONSTR:
	case STABLE_NAME:
	case BCO:
	case CONSTR_STATIC:
	    // StgMutArrPtr.ptrs, no SRT
894 895
	case MUT_ARR_PTRS_CLEAN:
	case MUT_ARR_PTRS_DIRTY:
896
	case MUT_ARR_PTRS_FROZEN:
897
	case MUT_ARR_PTRS_FROZEN0:
898 899 900 901 902 903 904 905 906 907 908 909
	    *c = find_ptrs(&se->info);
	    if (*c == NULL) {
		popOff();
		break;
	    }
	    *cp = se->c;
	    *r = se->c_child_r;
	    return;

	    // layout.payload.ptrs, SRT
	case FUN:         // always a heap object
	case FUN_2_0:
910 911 912 913 914 915 916 917 918 919 920
	    if (se->info.type == posTypePtrs) {
		*c = find_ptrs(&se->info);
		if (*c != NULL) {
		    *cp = se->c;
		    *r = se->c_child_r;
		    return;
		}
		init_srt_fun(&se->info, get_fun_itbl(se->c));
	    }
	    goto do_srt;

921 922 923 924 925 926 927 928 929
	case THUNK:
	case THUNK_2_0:
	    if (se->info.type == posTypePtrs) {
		*c = find_ptrs(&se->info);
		if (*c != NULL) {
		    *cp = se->c;
		    *r = se->c_child_r;
		    return;
		}
930
		init_srt_thunk(&se->info, get_thunk_itbl(se->c));
931
	    }
932
	    goto do_srt;
933 934

	    // SRT
935
	do_srt:
936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963
	case THUNK_STATIC:
	case FUN_STATIC:
	case FUN_0_1:
	case FUN_0_2:
	case THUNK_0_1:
	case THUNK_0_2:
	case FUN_1_0:
	case FUN_1_1:
	case THUNK_1_0:
	case THUNK_1_1:
	    *c = find_srt(&se->info);
	    if (*c != NULL) {
		*cp = se->c;
		*r = se->c_child_r;
		return;
	    }
	    popOff();
	    break;

	    // no child (fixed), no SRT
	case CONSTR_0_1:
	case CONSTR_0_2:
	case CAF_BLACKHOLE:
	case BLACKHOLE:
	case SE_BLACKHOLE:
	case SE_CAF_BLACKHOLE:
	case ARR_WORDS:
	    // one child (fixed), no SRT
964 965
	case MUT_VAR_CLEAN:
	case MUT_VAR_DIRTY:
966 967 968 969 970 971 972
	case THUNK_SELECTOR:
	case IND_PERM:
	case IND_OLDGEN_PERM:
	case IND_OLDGEN:
	case CONSTR_1_1:
	    // cannot appear
	case PAP:
973 974
	case AP:
	case AP_STACK:
975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041
	case TSO:
	case IND_STATIC:
	case CONSTR_INTLIKE:
	case CONSTR_CHARLIKE:
	case CONSTR_NOCAF_STATIC:
	    // stack objects
	case RET_DYN:
	case UPDATE_FRAME:
	case CATCH_FRAME:
	case STOP_FRAME:
	case RET_BCO:
	case RET_SMALL:
	case RET_VEC_SMALL:
	case RET_BIG:
	case RET_VEC_BIG:
	    // invalid objects
	case IND:
	case BLOCKED_FETCH:
	case FETCH_ME:
	case FETCH_ME_BQ:
	case RBH:
	case REMOTE_REF:
	case EVACUATED:
	case INVALID_OBJECT:
	default:
	    barf("Invalid object *c in pop()");
	    return;
	}
    } while (rtsTrue);
}

/* -----------------------------------------------------------------------------
 * RETAINER PROFILING ENGINE
 * -------------------------------------------------------------------------- */

void
initRetainerProfiling( void )
{
    initializeAllRetainerSet();
    retainerGeneration = 0;
}

/* -----------------------------------------------------------------------------
 *  This function must be called before f-closing prof_file.
 * -------------------------------------------------------------------------- */
void
endRetainerProfiling( void )
{
#ifdef SECOND_APPROACH
    outputAllRetainerSet(prof_file);
#endif
}

/* -----------------------------------------------------------------------------
 *  Returns the actual pointer to the retainer set of the closure *c.
 *  It may adjust RSET(c) subject to flip.
 *  Side effects:
 *    RSET(c) is initialized to NULL if its current value does not
 *    conform to flip.
 *  Note:
 *    Even though this function has side effects, they CAN be ignored because
 *    subsequent calls to retainerSetOf() always result in the same return value
 *    and retainerSetOf() is the only way to retrieve retainerSet of a given
 *    closure.
 *    We have to perform an XOR (^) operation each time a closure is examined.
 *    The reason is that we do not know when a closure is visited last.
 * -------------------------------------------------------------------------- */
1042
static INLINE void
1043 1044 1045 1046 1047 1048 1049 1050 1051 1052
maybeInitRetainerSet( StgClosure *c )
{
    if (!isRetainerSetFieldValid(c)) {
	setRetainerSetToNull(c);
    }
}

/* -----------------------------------------------------------------------------
 * Returns rtsTrue if *c is a retainer.
 * -------------------------------------------------------------------------- */
1053
static INLINE rtsBool
1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064
isRetainer( StgClosure *c )
{
    switch (get_itbl(c)->type) {
	//
	//  True case
	//
	// TSOs MUST be retainers: they constitute the set of roots.
    case TSO:

	// mutable objects
    case MVAR:
1065 1066
    case MUT_VAR_CLEAN:
    case MUT_VAR_DIRTY:
1067 1068
    case MUT_ARR_PTRS_CLEAN:
    case MUT_ARR_PTRS_DIRTY:
1069
    case MUT_ARR_PTRS_FROZEN:
1070
    case MUT_ARR_PTRS_FROZEN0:
1071 1072 1073 1074 1075 1076 1077 1078 1079

	// thunks are retainers.
    case THUNK:
    case THUNK_1_0:
    case THUNK_0_1:
    case THUNK_2_0:
    case THUNK_1_1:
    case THUNK_0_2:
    case THUNK_SELECTOR:
1080 1081
    case AP:
    case AP_STACK:
1082 1083 1084 1085 1086 1087 1088

	// Static thunks, or CAFS, are obviously retainers.
    case THUNK_STATIC:

	// WEAK objects are roots; there is separate code in which traversing
	// begins from WEAK objects.
    case WEAK:
1089 1090 1091 1092

	// Since the other mutvar-type things are retainers, seems
	// like the right thing to do:
    case TVAR:
1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130
	return rtsTrue;

	//
	// False case
	//

	// constructors
    case CONSTR:
    case CONSTR_1_0:
    case CONSTR_0_1:
    case CONSTR_2_0:
    case CONSTR_1_1:
    case CONSTR_0_2:
	// functions
    case FUN:
    case FUN_1_0:
    case FUN_0_1:
    case FUN_2_0:
    case FUN_1_1:
    case FUN_0_2:
	// partial applications
    case PAP:
	// blackholes
    case CAF_BLACKHOLE:
    case BLACKHOLE:
    case SE_BLACKHOLE:
    case SE_CAF_BLACKHOLE:
	// indirection
    case IND_PERM:
    case IND_OLDGEN_PERM:
    case IND_OLDGEN:
	// static objects
    case CONSTR_STATIC:
    case FUN_STATIC:
	// misc
    case STABLE_NAME:
    case BCO:
    case ARR_WORDS:
1131 1132 1133 1134
	// STM
    case TVAR_WAIT_QUEUE:
    case TREC_HEADER:
    case TREC_CHUNK:
1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181
	return rtsFalse;

	//
	// Error case
	//
	// IND_STATIC cannot be *c, *cp, *r in the retainer profiling loop.
    case IND_STATIC:
	// CONSTR_INTLIKE, CONSTR_CHARLIKE, and CONSTR_NOCAF_STATIC
	// cannot be *c, *cp, *r in the retainer profiling loop.
    case CONSTR_INTLIKE:
    case CONSTR_CHARLIKE:
    case CONSTR_NOCAF_STATIC:
	// Stack objects are invalid because they are never treated as
	// legal objects during retainer profiling.
    case UPDATE_FRAME:
    case CATCH_FRAME:
    case STOP_FRAME:
    case RET_DYN:
    case RET_BCO:
    case RET_SMALL:
    case RET_VEC_SMALL:
    case RET_BIG:
    case RET_VEC_BIG:
	// other cases
    case IND:
    case BLOCKED_FETCH:
    case FETCH_ME:
    case FETCH_ME_BQ:
    case RBH:
    case REMOTE_REF:
    case EVACUATED:
    case INVALID_OBJECT:
    default:
	barf("Invalid object in isRetainer(): %d", get_itbl(c)->type);
	return rtsFalse;
    }
}

/* -----------------------------------------------------------------------------
 *  Returns the retainer function value for the closure *c, i.e., R(*c).
 *  This function does NOT return the retainer(s) of *c.
 *  Invariants:
 *    *c must be a retainer.
 *  Note:
 *    Depending on the definition of this function, the maintenance of retainer
 *    sets can be made easier. If most retainer sets are likely to be created
 *    again across garbage collections, refreshAllRetainerSet() in
1182
 *    RetainerSet.c can simply do nothing.
1183 1184 1185 1186
 *    If this is not the case, we can free all the retainer sets and
 *    re-initialize the hash table.
 *    See refreshAllRetainerSet() in RetainerSet.c.
 * -------------------------------------------------------------------------- */
1187
static INLINE retainer
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210
getRetainerFrom( StgClosure *c )
{
    ASSERT(isRetainer(c));

#if defined(RETAINER_SCHEME_INFO)
    // Retainer scheme 1: retainer = info table
    return get_itbl(c);
#elif defined(RETAINER_SCHEME_CCS)
    // Retainer scheme 2: retainer = cost centre stack
    return c->header.prof.ccs;
#elif defined(RETAINER_SCHEME_CC)
    // Retainer scheme 3: retainer = cost centre
    return c->header.prof.ccs->cc;
#endif
}

/* -----------------------------------------------------------------------------
 *  Associates the retainer set *s with the closure *c, that is, *s becomes
 *  the retainer set of *c.
 *  Invariants:
 *    c != NULL
 *    s != NULL
 * -------------------------------------------------------------------------- */
1211
static INLINE void
1212
associate( StgClosure *c, RetainerSet *s )
1213 1214 1215 1216 1217 1218
{
    // StgWord has the same size as pointers, so the following type
    // casting is okay.
    RSET(c) = (RetainerSet *)((StgWord)s | flip);
}

1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246
/* -----------------------------------------------------------------------------
   Call retainClosure for each of the closures covered by a large bitmap.
   -------------------------------------------------------------------------- */

static void
retain_large_bitmap (StgPtr p, StgLargeBitmap *large_bitmap, nat size,
		     StgClosure *c, retainer c_child_r)
{
    nat i, b;
    StgWord bitmap;
    
    b = 0;
    bitmap = large_bitmap->bitmap[b];
    for (i = 0; i < size; ) {
	if ((bitmap & 1) == 0) {
	    retainClosure((StgClosure *)*p, c, c_child_r);
	}
	i++;
	p++;
	if (i % BITS_IN(W_) == 0) {
	    b++;
	    bitmap = large_bitmap->bitmap[b];
	} else {
	    bitmap = bitmap >> 1;
	}
    }
}

1247
static INLINE StgPtr
1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261
retain_small_bitmap (StgPtr p, nat size, StgWord bitmap,
		     StgClosure *c, retainer c_child_r)
{
    while (size > 0) {
	if ((bitmap & 1) == 0) {
	    retainClosure((StgClosure *)*p, c, c_child_r);
	}
	p++;
	bitmap = bitmap >> 1;
	size--;
    }
    return p;
}

1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323
/* -----------------------------------------------------------------------------
 * Call retainClosure for each of the closures in an SRT.
 * ------------------------------------------------------------------------- */

static void
retain_large_srt_bitmap (StgLargeSRT *srt, StgClosure *c, retainer c_child_r)
{
    nat i, b, size;
    StgWord bitmap;
    StgClosure **p;
    
    b = 0;
    p = (StgClosure **)srt->srt;
    size   = srt->l.size;
    bitmap = srt->l.bitmap[b];
    for (i = 0; i < size; ) {
	if ((bitmap & 1) != 0) {
	    retainClosure((StgClosure *)*p, c, c_child_r);
	}
	i++;
	p++;
	if (i % BITS_IN(W_) == 0) {
	    b++;
	    bitmap = srt->l.bitmap[b];
	} else {
	    bitmap = bitmap >> 1;
	}
    }
}

static INLINE void
retainSRT (StgClosure **srt, nat srt_bitmap, StgClosure *c, retainer c_child_r)
{
  nat bitmap;
  StgClosure **p;

  bitmap = srt_bitmap;
  p = srt;

  if (bitmap == (StgHalfWord)(-1)) {  
      retain_large_srt_bitmap( (StgLargeSRT *)srt, c, c_child_r );
      return;
  }

  while (bitmap != 0) {
      if ((bitmap & 1) != 0) {
#ifdef ENABLE_WIN32_DLL_SUPPORT
	  if ( (unsigned long)(*srt) & 0x1 ) {
	      retainClosure(*stgCast(StgClosure**,(stgCast(unsigned long, *srt) & ~0x1)), 
			    c, c_child_r);
	  } else {
	      retainClosure(*srt,c,c_child_r);
	  }
#else
	  retainClosure(*srt,c,c_child_r);
#endif
      }
      p++;
      bitmap = bitmap >> 1;
  }
}

1324 1325 1326 1327 1328 1329
/* -----------------------------------------------------------------------------
 *  Process all the objects in the stack chunk from stackStart to stackEnd
 *  with *c and *c_child_r being their parent and their most recent retainer,
 *  respectively. Treat stackOptionalFun as another child of *c if it is
 *  not NULL.
 *  Invariants:
1330
 *    *c is one of the following: TSO, AP_STACK.
1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342
 *    If *c is TSO, c == c_child_r.
 *    stackStart < stackEnd.
 *    RSET(c) and RSET(c_child_r) are valid, i.e., their
 *    interpretation conforms to the current value of flip (even when they
 *    are interpreted to be NULL).
 *    If *c is TSO, its state is not any of ThreadRelocated, ThreadComplete,
 *    or ThreadKilled, which means that its stack is ready to process.
 *  Note:
 *    This code was almost plagiarzied from GC.c! For each pointer,
 *    retainClosure() is invoked instead of evacuate().
 * -------------------------------------------------------------------------- */
static void
1343
retainStack( StgClosure *c, retainer c_child_r,
1344
	     StgPtr stackStart, StgPtr stackEnd )
1345 1346
{
    stackElement *oldStackBoundary;
1347 1348
    StgPtr p;
    StgRetInfoTable *info;
1349
    StgWord32 bitmap;
1350
    nat size;
1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366

#ifdef DEBUG_RETAINER
    cStackSize++;
    if (cStackSize > maxCStackSize) maxCStackSize = cStackSize;
#endif

    /*
      Each invocation of retainStack() creates a new virtual
      stack. Since all such stacks share a single common stack, we
      record the current currentStackBoundary, which will be restored
      at the exit.
    */
    oldStackBoundary = currentStackBoundary;
    currentStackBoundary = stackTop;

#ifdef DEBUG_RETAINER
1367
    // debugBelch("retainStack() called: oldStackBoundary = 0x%x, currentStackBoundary = 0x%x\n", oldStackBoundary, currentStackBoundary);
1368 1369
#endif

1370 1371 1372 1373 1374
    ASSERT(get_itbl(c)->type != TSO || 
	   (((StgTSO *)c)->what_next != ThreadRelocated &&
	    ((StgTSO *)c)->what_next != ThreadComplete &&
	    ((StgTSO *)c)->what_next != ThreadKilled));
    
1375 1376
    p = stackStart;
    while (p < stackEnd) {
1377
	info = get_ret_itbl((StgClosure *)p);
1378

1379
	switch(info->i.type) {
1380 1381 1382 1383 1384 1385 1386 1387

	case UPDATE_FRAME:
	    retainClosure(((StgUpdateFrame *)p)->updatee, c, c_child_r);
	    p += sizeofW(StgUpdateFrame);
	    continue;

	case STOP_FRAME:
	case CATCH_FRAME:
1388 1389 1390
	case CATCH_STM_FRAME:
	case CATCH_RETRY_FRAME:
	case ATOMICALLY_FRAME:
1391 1392
	case RET_SMALL:
	case RET_VEC_SMALL:
1393 1394
	    bitmap = BITMAP_BITS(info->i.layout.bitmap);
	    size   = BITMAP_SIZE(info->i.layout.bitmap);
1395
	    p++;
1396 1397
	    p = retain_small_bitmap(p, size, bitmap, c, c_child_r);

1398
	follow_srt:
1399
	    retainSRT((StgClosure **)GET_SRT(info), info->i.srt_bitmap, c, c_child_r);
1400
	    continue;
1401

1402 1403 1404 1405 1406 1407 1408 1409 1410 1411
	case RET_BCO: {
	    StgBCO *bco;
	    
	    p++;
	    retainClosure((StgClosure *)*p, c, c_child_r);
	    bco = (StgBCO *)*p;
	    p++;
	    size = BCO_BITMAP_SIZE(bco);
	    retain_large_bitmap(p, BCO_BITMAP(bco), size, c, c_child_r);
	    p += size;
1412
	    continue;
1413
	}
1414

1415
	    // large bitmap (> 32 entries, or > 64 on a 64-bit machine) 
1416 1417
	case RET_BIG:
	case RET_VEC_BIG:
1418
	    size = GET_LARGE_BITMAP(&info->i)->size;
1419
	    p++;
1420
	    retain_large_bitmap(p, GET_LARGE_BITMAP(&info->i),
1421 1422 1423 1424
				size, c, c_child_r);
	    p += size;
	    // and don't forget to follow the SRT 
	    goto follow_srt;
1425

1426 1427 1428 1429 1430 1431
	    // Dynamic bitmap: the mask is stored on the stack 
	case RET_DYN: {
	    StgWord dyn;
	    dyn = ((StgRetDyn *)p)->liveness;

	    // traverse the bitmap first
1432
	    bitmap = RET_DYN_LIVENESS(dyn);
1433
	    p      = (P_)&((StgRetDyn *)p)->payload[0];
1434
	    size   = RET_DYN_BITMAP_SIZE;
1435 1436 1437
	    p = retain_small_bitmap(p, size, bitmap, c, c_child_r);
	    
	    // skip over the non-ptr words
1438
	    p += RET_DYN_NONPTRS(dyn) + RET_DYN_NONPTR_REGS_SIZE;
1439 1440
	    
	    // follow the ptr words
1441
	    for (size = RET_DYN_PTRS(dyn); size > 0; size--) {
1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455
		retainClosure((StgClosure *)*p, c, c_child_r);
		p++;
	    }
	    continue;
	}

	case RET_FUN: {
	    StgRetFun *ret_fun = (StgRetFun *)p;
	    StgFunInfoTable *fun_info;
	    
	    retainClosure(ret_fun->fun, c, c_child_r);
	    fun_info = get_fun_itbl(ret_fun->fun);
	    
	    p = (P_)&ret_fun->payload;
1456
	    switch (fun_info->f.fun_type) {
1457
	    case ARG_GEN:
1458 1459
		bitmap = BITMAP_BITS(fun_info->f.b.bitmap);
		size = BITMAP_SIZE(fun_info->f.b.bitmap);
1460 1461 1462
		p = retain_small_bitmap(p, size, bitmap, c, c_child_r);
		break;
	    case ARG_GEN_BIG:
1463 1464
		size = GET_FUN_LARGE_BITMAP(fun_info)->size;
		retain_large_bitmap(p, GET_FUN_LARGE_BITMAP(fun_info), 
1465 1466 1467 1468
				    size, c, c_child_r);
		p += size;
		break;
	    default:
1469 1470
		bitmap = BITMAP_BITS(stg_arg_bitmaps[fun_info->f.fun_type]);
		size = BITMAP_SIZE(stg_arg_bitmaps[fun_info->f.fun_type]);
1471 1472
		p = retain_small_bitmap(p, size, bitmap, c, c_child_r);
		break;
1473
	    }
1474
	    goto follow_srt;
1475
	}
1476

1477 1478
	default:
	    barf("Invalid object found in retainStack(): %d",
1479
		 (int)(info->i.type));
1480 1481 1482 1483 1484 1485
	}
    }

    // restore currentStackBoundary
    currentStackBoundary = oldStackBoundary;
#ifdef DEBUG_RETAINER
1486
    // debugBelch("retainStack() finished: currentStackBoundary = 0x%x\n", currentStackBoundary);
1487 1488 1489 1490 1491 1492 1493
#endif

#ifdef DEBUG_RETAINER
    cStackSize--;
#endif
}

1494 1495 1496 1497
/* ----------------------------------------------------------------------------
 * Call retainClosure for each of the children of a PAP/AP
 * ------------------------------------------------------------------------- */

1498
static INLINE StgPtr
1499 1500
retain_PAP_payload (StgClosure *pap,  retainer c_child_r, StgClosure *fun, 
		    StgClosure** payload, StgWord n_args)
1501 1502
{
    StgPtr p;
1503
    StgWord bitmap;
1504 1505
    StgFunInfoTable *fun_info;

1506 1507
    retainClosure(fun, pap, c_child_r);
    fun_info = get_fun_itbl(fun);
1508 1509
    ASSERT(fun_info->i.type != PAP);

1510
    p = (StgPtr)payload;
1511

1512
    switch (fun_info->f.fun_type) {
1513
    case ARG_GEN:
1514
	bitmap = BITMAP_BITS(fun_info->f.b.bitmap);
1515 1516
	p = retain_small_bitmap(p, n_args, bitmap, 
				pap, c_child_r);
1517 1518
	break;
    case ARG_GEN_BIG:
1519
	retain_large_bitmap(p, GET_FUN_LARGE_BITMAP(fun_info),
1520 1521
			    n_args, pap, c_child_r);
	p += n_args;
1522 1523
	break;
    case ARG_BCO:
1524 1525 1526
	retain_large_bitmap((StgPtr)payload, BCO_BITMAP(fun),
			    n_args, pap, c_child_r);
	p += n_args;
1527 1528
	break;
    default:
1529
	bitmap = BITMAP_BITS(stg_arg_bitmaps[fun_info->f.fun_type]);
1530
	p = retain_small_bitmap(p, n_args, bitmap, pap, c_child_r);
1531 1532 1533 1534 1535
	break;
    }
    return p;
}

1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549
/* -----------------------------------------------------------------------------
 *  Compute the retainer set of *c0 and all its desecents by traversing.
 *  *cp0 is the parent of *c0, and *r0 is the most recent retainer of *c0.
 *  Invariants:
 *    c0 = cp0 = r0 holds only for root objects.
 *    RSET(cp0) and RSET(r0) are valid, i.e., their
 *    interpretation conforms to the current value of flip (even when they
 *    are interpreted to be NULL).
 *    However, RSET(c0) may be corrupt, i.e., it may not conform to
 *    the current value of flip. If it does not, during the execution
 *    of this function, RSET(c0) must be initialized as well as all
 *    its descendants.
 *  Note:
 *    stackTop must be the same at the beginning and the exit of this function.
1550
 *    *c0 can be TSO (as well as AP_STACK).
1551 1552
 * -------------------------------------------------------------------------- */
static void
1553
retainClosure( StgClosure *c0, StgClosure *cp0, retainer r0 )
1554 1555 1556 1557 1558 1559
{
    // c = Current closure
    // cp = Current closure's Parent
    // r = current closures' most recent Retainer
    // c_child_r = current closure's children's most recent retainer
    // first_child = first child of c
1560
    StgClosure *c, *cp, *first_child;
1561
    RetainerSet *s, *retainerSetOfc;
1562
    retainer r, c_child_r;
1563 1564 1565 1566 1567 1568 <