1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2004
5 * Out-of-line primitive operations
7 * This file contains the implementations of all the primitive
8 * operations ("primops") which are not expanded inline. See
9 * ghc/compiler/prelude/primops.txt.pp for a list of all the primops;
10 * this file contains code for most of those with the attribute
13 * Entry convention: the entry convention for a primop is that all the
14 * args are in Stg registers (R1, R2, etc.). This is to make writing
15 * the primops easier. (see compiler/codeGen/CgCallConv.hs).
17 * Return convention: results from a primop are generally returned
18 * using the ordinary unboxed tuple return convention. The C-- parser
19 * implements the RET_xxxx() macros to perform unboxed-tuple returns
20 * based on the prevailing return convention.
22 * This file is written in a subset of C--, extended with various
23 * features specific to GHC. It is compiled by GHC directly. For the
24 * syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
26 * ---------------------------------------------------------------------------*/
31 import pthread_mutex_lock;
32 import pthread_mutex_unlock;
34 import base_ControlziExceptionziBase_nestedAtomically_closure;
35 import EnterCriticalSection;
36 import LeaveCriticalSection;
37 import ghczmprim_GHCziBool_False_closure;
39 /*-----------------------------------------------------------------------------
42 Basically just new*Array - the others are all inline macros.
44 The size arg is always passed in R1, and the result returned in R1.
46 The slow entry point is for returning from a heap check, the saved
47 size argument must be re-loaded from the stack.
48 -------------------------------------------------------------------------- */
50 /* for objects that are *less* than the size of a word, make sure we
51 * round up to the nearest word for the size of the array.
56 W_ words, payload_words, n, p;
57 MAYBE_GC(NO_PTRS,stg_newByteArrayzh);
59 payload_words = ROUNDUP_BYTES_TO_WDS(n);
60 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
61 ("ptr" p) = foreign "C" allocate(MyCapability() "ptr",words) [];
62 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
63 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
64 StgArrWords_words(p) = payload_words;
69 #define BA_MASK (BA_ALIGN-1)
71 stg_newPinnedByteArrayzh
73 W_ words, bytes, payload_words, p;
75 MAYBE_GC(NO_PTRS,stg_newPinnedByteArrayzh);
77 /* payload_words is what we will tell the profiler we had to allocate */
78 payload_words = ROUNDUP_BYTES_TO_WDS(bytes);
79 /* When we actually allocate memory, we need to allow space for the
81 bytes = bytes + SIZEOF_StgArrWords;
82 /* And we want to align to BA_ALIGN bytes, so we need to allow space
83 to shift up to BA_ALIGN - 1 bytes: */
84 bytes = bytes + BA_ALIGN - 1;
85 /* Now we convert to a number of words: */
86 words = ROUNDUP_BYTES_TO_WDS(bytes);
88 ("ptr" p) = foreign "C" allocatePinned(MyCapability() "ptr", words) [];
89 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
91 /* Now we need to move p forward so that the payload is aligned
93 p = p + ((-p - SIZEOF_StgArrWords) & BA_MASK);
95 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
96 StgArrWords_words(p) = payload_words;
100 stg_newAlignedPinnedByteArrayzh
102 W_ words, bytes, payload_words, p, alignment;
104 MAYBE_GC(NO_PTRS,stg_newAlignedPinnedByteArrayzh);
108 /* payload_words is what we will tell the profiler we had to allocate */
109 payload_words = ROUNDUP_BYTES_TO_WDS(bytes);
111 /* When we actually allocate memory, we need to allow space for the
113 bytes = bytes + SIZEOF_StgArrWords;
114 /* And we want to align to <alignment> bytes, so we need to allow space
115 to shift up to <alignment - 1> bytes: */
116 bytes = bytes + alignment - 1;
117 /* Now we convert to a number of words: */
118 words = ROUNDUP_BYTES_TO_WDS(bytes);
120 ("ptr" p) = foreign "C" allocatePinned(MyCapability() "ptr", words) [];
121 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
123 /* Now we need to move p forward so that the payload is aligned
124 to <alignment> bytes. Note that we are assuming that
125 <alignment> is a power of 2, which is technically not guaranteed */
126 p = p + ((-p - SIZEOF_StgArrWords) & (alignment - 1));
128 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
129 StgArrWords_words(p) = payload_words;
135 W_ words, n, init, arr, p;
136 /* Args: R1 = words, R2 = initialisation value */
139 MAYBE_GC(R2_PTR,stg_newArrayzh);
141 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
142 ("ptr" arr) = foreign "C" allocate(MyCapability() "ptr",words) [R2];
143 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
145 SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, W_[CCCS]);
146 StgMutArrPtrs_ptrs(arr) = n;
148 // Initialise all elements of the the array with the value in R2
150 p = arr + SIZEOF_StgMutArrPtrs;
152 if (p < arr + WDS(words)) {
161 stg_unsafeThawArrayzh
163 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
165 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
166 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
167 // it on the mutable list for the GC to remove (removing something from
168 // the mutable list is not easy, because the mut_list is only singly-linked).
170 // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list,
171 // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0
172 // to indicate that it is still on the mutable list.
174 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
175 // either it is on a mut_list, or it isn't. We adopt the convention that
176 // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list,
177 // and MUT_ARR_PTRS_FROZEN otherwise. In fact it wouldn't matter if
178 // we put it on the mutable list more than once, but it would get scavenged
179 // multiple times during GC, which would be unnecessarily slow.
181 if (StgHeader_info(R1) != stg_MUT_ARR_PTRS_FROZEN0_info) {
182 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
183 recordMutable(R1, R1);
184 // must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
187 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
192 /* -----------------------------------------------------------------------------
194 -------------------------------------------------------------------------- */
199 /* Args: R1 = initialisation value */
201 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, stg_newMutVarzh);
203 mv = Hp - SIZEOF_StgMutVar + WDS(1);
204 SET_HDR(mv,stg_MUT_VAR_DIRTY_info,W_[CCCS]);
205 StgMutVar_var(mv) = R1;
210 stg_atomicModifyMutVarzh
212 W_ mv, f, z, x, y, r, h;
213 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
215 /* If x is the current contents of the MutVar#, then
216 We want to make the new contents point to
220 and the return value is
224 obviously we can share (f x).
226 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
227 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
228 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
232 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
233 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
235 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
236 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
240 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
241 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
243 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
244 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
247 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
249 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, stg_atomicModifyMutVarzh);
254 TICK_ALLOC_THUNK_2();
255 CCCS_ALLOC(THUNK_2_SIZE);
256 z = Hp - THUNK_2_SIZE + WDS(1);
257 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
258 LDV_RECORD_CREATE(z);
259 StgThunk_payload(z,0) = f;
261 TICK_ALLOC_THUNK_1();
262 CCCS_ALLOC(THUNK_1_SIZE);
263 y = z - THUNK_1_SIZE;
264 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
265 LDV_RECORD_CREATE(y);
266 StgThunk_payload(y,0) = z;
268 TICK_ALLOC_THUNK_1();
269 CCCS_ALLOC(THUNK_1_SIZE);
270 r = y - THUNK_1_SIZE;
271 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
272 LDV_RECORD_CREATE(r);
273 StgThunk_payload(r,0) = z;
276 x = StgMutVar_var(mv);
277 StgThunk_payload(z,1) = x;
279 (h) = foreign "C" cas(mv + SIZEOF_StgHeader + OFFSET_StgMutVar_var, x, y) [];
280 if (h != x) { goto retry; }
282 StgMutVar_var(mv) = y;
285 if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) {
286 foreign "C" dirty_MUT_VAR(BaseReg "ptr", mv "ptr") [];
292 /* -----------------------------------------------------------------------------
293 Weak Pointer Primitives
294 -------------------------------------------------------------------------- */
296 STRING(stg_weak_msg,"New weak pointer at %p\n")
302 R3 = finalizer (or NULL)
307 R3 = stg_NO_FINALIZER_closure;
310 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, stg_mkWeakzh );
312 w = Hp - SIZEOF_StgWeak + WDS(1);
313 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
315 // We don't care about cfinalizer here.
316 // Should StgWeak_cfinalizer(w) be stg_NO_FINALIZER_closure or
320 StgWeak_value(w) = R2;
321 StgWeak_finalizer(w) = R3;
322 StgWeak_cfinalizer(w) = stg_NO_FINALIZER_closure;
324 ACQUIRE_LOCK(sm_mutex);
325 StgWeak_link(w) = W_[weak_ptr_list];
326 W_[weak_ptr_list] = w;
327 RELEASE_LOCK(sm_mutex);
329 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
334 stg_mkWeakForeignEnvzh
340 R5 = has environment (0 or 1)
343 W_ w, payload_words, words, p;
345 W_ key, val, fptr, ptr, flag, eptr;
354 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR, stg_mkWeakForeignEnvzh );
356 w = Hp - SIZEOF_StgWeak + WDS(1);
357 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
360 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
361 ("ptr" p) = foreign "C" allocate(MyCapability() "ptr", words) [];
363 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
364 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
366 StgArrWords_words(p) = payload_words;
367 StgArrWords_payload(p,0) = fptr;
368 StgArrWords_payload(p,1) = ptr;
369 StgArrWords_payload(p,2) = eptr;
370 StgArrWords_payload(p,3) = flag;
372 // We don't care about the value here.
373 // Should StgWeak_value(w) be stg_NO_FINALIZER_closure or something else?
375 StgWeak_key(w) = key;
376 StgWeak_value(w) = val;
377 StgWeak_finalizer(w) = stg_NO_FINALIZER_closure;
378 StgWeak_cfinalizer(w) = p;
380 StgWeak_link(w) = W_[weak_ptr_list];
381 W_[weak_ptr_list] = w;
383 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
397 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
398 RET_NP(0,stg_NO_FINALIZER_closure);
404 // A weak pointer is inherently used, so we do not need to call
405 // LDV_recordDead_FILL_SLOP_DYNAMIC():
406 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
407 // or, LDV_recordDead():
408 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
409 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
410 // large as weak pointers, so there is no need to fill the slop, either.
411 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
415 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
417 SET_INFO(w,stg_DEAD_WEAK_info);
418 LDV_RECORD_CREATE(w);
420 f = StgWeak_finalizer(w);
421 arr = StgWeak_cfinalizer(w);
423 StgDeadWeak_link(w) = StgWeak_link(w);
425 if (arr != stg_NO_FINALIZER_closure) {
426 foreign "C" runCFinalizer(StgArrWords_payload(arr,0),
427 StgArrWords_payload(arr,1),
428 StgArrWords_payload(arr,2),
429 StgArrWords_payload(arr,3)) [];
432 /* return the finalizer */
433 if (f == stg_NO_FINALIZER_closure) {
434 RET_NP(0,stg_NO_FINALIZER_closure);
446 if (GET_INFO(w) == stg_WEAK_info) {
448 val = StgWeak_value(w);
456 /* -----------------------------------------------------------------------------
457 Floating point operations.
458 -------------------------------------------------------------------------- */
460 stg_decodeFloatzuIntzh
467 STK_CHK_GEN( WDS(2), NO_PTRS, stg_decodeFloatzuIntzh );
469 mp_tmp1 = Sp - WDS(1);
470 mp_tmp_w = Sp - WDS(2);
472 /* arguments: F1 = Float# */
475 /* Perform the operation */
476 foreign "C" __decodeFloat_Int(mp_tmp1 "ptr", mp_tmp_w "ptr", arg) [];
478 /* returns: (Int# (mantissa), Int# (exponent)) */
479 RET_NN(W_[mp_tmp1], W_[mp_tmp_w]);
482 stg_decodeDoublezu2Intzh
491 STK_CHK_GEN( WDS(4), NO_PTRS, stg_decodeDoublezu2Intzh );
493 mp_tmp1 = Sp - WDS(1);
494 mp_tmp2 = Sp - WDS(2);
495 mp_result1 = Sp - WDS(3);
496 mp_result2 = Sp - WDS(4);
498 /* arguments: D1 = Double# */
501 /* Perform the operation */
502 foreign "C" __decodeDouble_2Int(mp_tmp1 "ptr", mp_tmp2 "ptr",
503 mp_result1 "ptr", mp_result2 "ptr",
507 (Int# (mant sign), Word# (mant high), Word# (mant low), Int# (expn)) */
508 RET_NNNN(W_[mp_tmp1], W_[mp_tmp2], W_[mp_result1], W_[mp_result2]);
511 /* -----------------------------------------------------------------------------
512 * Concurrency primitives
513 * -------------------------------------------------------------------------- */
517 /* args: R1 = closure to spark */
519 MAYBE_GC(R1_PTR, stg_forkzh);
525 ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr",
526 RtsFlags_GcFlags_initialStkSize(RtsFlags),
529 /* start blocked if the current thread is blocked */
530 StgTSO_flags(threadid) =
531 StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) &
532 (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));
534 foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") [];
536 // context switch soon, but not immediately: we don't want every
537 // forkIO to force a context-switch.
538 Capability_context_switch(MyCapability()) = 1 :: CInt;
545 /* args: R1 = cpu, R2 = closure to spark */
547 MAYBE_GC(R2_PTR, stg_forkOnzh);
555 ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr",
556 RtsFlags_GcFlags_initialStkSize(RtsFlags),
559 /* start blocked if the current thread is blocked */
560 StgTSO_flags(threadid) =
561 StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) &
562 (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));
564 foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") [];
566 // context switch soon, but not immediately: we don't want every
567 // forkIO to force a context-switch.
568 Capability_context_switch(MyCapability()) = 1 :: CInt;
575 jump stg_yield_noregs;
590 foreign "C" labelThread(R1 "ptr", R2 "ptr") [];
592 jump %ENTRY_CODE(Sp(0));
595 stg_isCurrentThreadBoundzh
599 (r) = foreign "C" isThreadBound(CurrentTSO) [];
605 /* args: R1 :: ThreadId# */
613 if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
614 tso = StgTSO__link(tso);
618 what_next = TO_W_(StgTSO_what_next(tso));
619 why_blocked = TO_W_(StgTSO_why_blocked(tso));
620 // Note: these two reads are not atomic, so they might end up
621 // being inconsistent. It doesn't matter, since we
622 // only return one or the other. If we wanted to return the
623 // contents of block_info too, then we'd have to do some synchronisation.
625 if (what_next == ThreadComplete) {
626 ret = 16; // NB. magic, matches up with GHC.Conc.threadStatus
628 if (what_next == ThreadKilled) {
637 /* -----------------------------------------------------------------------------
639 * -------------------------------------------------------------------------- */
643 // Catch retry frame ------------------------------------------------------------
645 INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME,
646 #if defined(PROFILING)
647 W_ unused1, W_ unused2,
649 W_ unused3, P_ unused4, P_ unused5)
651 W_ r, frame, trec, outer;
654 trec = StgTSO_trec(CurrentTSO);
655 outer = StgTRecHeader_enclosing_trec(trec);
656 (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
658 /* Succeeded (either first branch or second branch) */
659 StgTSO_trec(CurrentTSO) = outer;
660 Sp = Sp + SIZEOF_StgCatchRetryFrame;
661 jump %ENTRY_CODE(Sp(SP_OFF));
663 /* Did not commit: re-execute */
665 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
666 StgTSO_trec(CurrentTSO) = new_trec;
667 if (StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
668 R1 = StgCatchRetryFrame_alt_code(frame);
670 R1 = StgCatchRetryFrame_first_code(frame);
677 // Atomically frame ------------------------------------------------------------
679 INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME,
680 #if defined(PROFILING)
681 W_ unused1, W_ unused2,
683 P_ code, P_ next_invariant_to_check, P_ result)
685 W_ frame, trec, valid, next_invariant, q, outer;
688 trec = StgTSO_trec(CurrentTSO);
690 outer = StgTRecHeader_enclosing_trec(trec);
692 if (outer == NO_TREC) {
693 /* First time back at the atomically frame -- pick up invariants */
694 ("ptr" q) = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") [];
695 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
696 StgAtomicallyFrame_result(frame) = result;
699 /* Second/subsequent time back at the atomically frame -- abort the
700 * tx that's checking the invariant and move on to the next one */
701 StgTSO_trec(CurrentTSO) = outer;
702 q = StgAtomicallyFrame_next_invariant_to_check(frame);
703 StgInvariantCheckQueue_my_execution(q) = trec;
704 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
705 /* Don't free trec -- it's linked from q and will be stashed in the
706 * invariant if we eventually commit. */
707 q = StgInvariantCheckQueue_next_queue_entry(q);
708 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
712 q = StgAtomicallyFrame_next_invariant_to_check(frame);
714 if (q != END_INVARIANT_CHECK_QUEUE) {
715 /* We can't commit yet: another invariant to check */
716 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [];
717 StgTSO_trec(CurrentTSO) = trec;
719 next_invariant = StgInvariantCheckQueue_invariant(q);
720 R1 = StgAtomicInvariant_code(next_invariant);
725 /* We've got no more invariants to check, try to commit */
726 (valid) = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") [];
728 /* Transaction was valid: commit succeeded */
729 StgTSO_trec(CurrentTSO) = NO_TREC;
730 R1 = StgAtomicallyFrame_result(frame);
731 Sp = Sp + SIZEOF_StgAtomicallyFrame;
732 jump %ENTRY_CODE(Sp(SP_OFF));
734 /* Transaction was not valid: try again */
735 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
736 StgTSO_trec(CurrentTSO) = trec;
737 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
738 R1 = StgAtomicallyFrame_code(frame);
744 INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME,
745 #if defined(PROFILING)
746 W_ unused1, W_ unused2,
748 P_ code, P_ next_invariant_to_check, P_ result)
750 W_ frame, trec, valid;
754 /* The TSO is currently waiting: should we stop waiting? */
755 (valid) = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") [];
757 /* Previous attempt is still valid: no point trying again yet */
758 jump stg_block_noregs;
760 /* Previous attempt is no longer valid: try again */
761 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
762 StgTSO_trec(CurrentTSO) = trec;
763 StgHeader_info(frame) = stg_atomically_frame_info;
764 R1 = StgAtomicallyFrame_code(frame);
769 // STM catch frame --------------------------------------------------------------
773 /* Catch frames are very similar to update frames, but when entering
774 * one we just pop the frame off the stack and perform the correct
775 * kind of return to the activation record underneath us on the stack.
778 INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME,
779 #if defined(PROFILING)
780 W_ unused1, W_ unused2,
782 P_ unused3, P_ unused4)
784 W_ r, frame, trec, outer;
786 trec = StgTSO_trec(CurrentTSO);
787 outer = StgTRecHeader_enclosing_trec(trec);
788 (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
790 /* Commit succeeded */
791 StgTSO_trec(CurrentTSO) = outer;
792 Sp = Sp + SIZEOF_StgCatchSTMFrame;
797 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
798 StgTSO_trec(CurrentTSO) = new_trec;
799 R1 = StgCatchSTMFrame_code(frame);
805 // Primop definition ------------------------------------------------------------
813 // stmStartTransaction may allocate
814 MAYBE_GC (R1_PTR, stg_atomicallyzh);
816 /* Args: R1 = m :: STM a */
817 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, stg_atomicallyzh);
819 old_trec = StgTSO_trec(CurrentTSO);
821 /* Nested transactions are not allowed; raise an exception */
822 if (old_trec != NO_TREC) {
823 R1 = base_ControlziExceptionziBase_nestedAtomically_closure;
827 /* Set up the atomically frame */
828 Sp = Sp - SIZEOF_StgAtomicallyFrame;
831 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
832 StgAtomicallyFrame_code(frame) = R1;
833 StgAtomicallyFrame_result(frame) = NO_TREC;
834 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
836 /* Start the memory transcation */
837 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1];
838 StgTSO_trec(CurrentTSO) = new_trec;
840 /* Apply R1 to the realworld token */
849 /* Args: R1 :: STM a */
850 /* Args: R2 :: Exception -> STM a */
851 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, stg_catchSTMzh);
853 /* Set up the catch frame */
854 Sp = Sp - SIZEOF_StgCatchSTMFrame;
857 SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
858 StgCatchSTMFrame_handler(frame) = R2;
859 StgCatchSTMFrame_code(frame) = R1;
861 /* Start a nested transaction to run the body of the try block in */
864 cur_trec = StgTSO_trec(CurrentTSO);
865 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr");
866 StgTSO_trec(CurrentTSO) = new_trec;
868 /* Apply R1 to the realworld token */
879 // stmStartTransaction may allocate
880 MAYBE_GC (R1_PTR & R2_PTR, stg_catchRetryzh);
882 /* Args: R1 :: STM a */
883 /* Args: R2 :: STM a */
884 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, stg_catchRetryzh);
886 /* Start a nested transaction within which to run the first code */
887 trec = StgTSO_trec(CurrentTSO);
888 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2];
889 StgTSO_trec(CurrentTSO) = new_trec;
891 /* Set up the catch-retry frame */
892 Sp = Sp - SIZEOF_StgCatchRetryFrame;
895 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
896 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
897 StgCatchRetryFrame_first_code(frame) = R1;
898 StgCatchRetryFrame_alt_code(frame) = R2;
900 /* Apply R1 to the realworld token */
913 MAYBE_GC (NO_PTRS, stg_retryzh); // STM operations may allocate
915 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
917 StgTSO_sp(CurrentTSO) = Sp;
918 (frame_type) = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
919 Sp = StgTSO_sp(CurrentTSO);
921 trec = StgTSO_trec(CurrentTSO);
922 outer = StgTRecHeader_enclosing_trec(trec);
924 if (frame_type == CATCH_RETRY_FRAME) {
925 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
926 ASSERT(outer != NO_TREC);
927 // Abort the transaction attempting the current branch
928 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
929 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
930 if (!StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
931 // Retry in the first branch: try the alternative
932 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
933 StgTSO_trec(CurrentTSO) = trec;
934 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
935 R1 = StgCatchRetryFrame_alt_code(frame);
938 // Retry in the alternative code: propagate the retry
939 StgTSO_trec(CurrentTSO) = outer;
940 Sp = Sp + SIZEOF_StgCatchRetryFrame;
941 goto retry_pop_stack;
945 // We've reached the ATOMICALLY_FRAME: attempt to wait
946 ASSERT(frame_type == ATOMICALLY_FRAME);
947 if (outer != NO_TREC) {
948 // We called retry while checking invariants, so abort the current
949 // invariant check (merging its TVar accesses into the parents read
950 // set so we'll wait on them)
951 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
952 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
954 StgTSO_trec(CurrentTSO) = trec;
955 outer = StgTRecHeader_enclosing_trec(trec);
957 ASSERT(outer == NO_TREC);
959 (r) = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") [];
961 // Transaction was valid: stmWait put us on the TVars' queues, we now block
962 StgHeader_info(frame) = stg_atomically_waiting_frame_info;
964 // Fix up the stack in the unregisterised case: the return convention is different.
965 R3 = trec; // passing to stmWaitUnblock()
966 jump stg_block_stmwait;
968 // Transaction was not valid: retry immediately
969 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
970 StgTSO_trec(CurrentTSO) = trec;
971 R1 = StgAtomicallyFrame_code(frame);
982 /* Args: R1 = invariant closure */
983 MAYBE_GC (R1_PTR, stg_checkzh);
985 trec = StgTSO_trec(CurrentTSO);
987 foreign "C" stmAddInvariantToCheck(MyCapability() "ptr",
991 jump %ENTRY_CODE(Sp(0));
1000 /* Args: R1 = initialisation value */
1002 MAYBE_GC (R1_PTR, stg_newTVarzh);
1004 ("ptr" tv) = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") [];
1015 /* Args: R1 = TVar closure */
1017 MAYBE_GC (R1_PTR, stg_readTVarzh); // Call to stmReadTVar may allocate
1018 trec = StgTSO_trec(CurrentTSO);
1020 ("ptr" result) = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1030 result = StgTVar_current_value(R1);
1031 if (%INFO_PTR(result) == stg_TREC_HEADER_info) {
1043 /* Args: R1 = TVar closure */
1044 /* R2 = New value */
1046 MAYBE_GC (R1_PTR & R2_PTR, stg_writeTVarzh); // Call to stmWriteTVar may allocate
1047 trec = StgTSO_trec(CurrentTSO);
1050 foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1052 jump %ENTRY_CODE(Sp(0));
1056 /* -----------------------------------------------------------------------------
1059 * take & putMVar work as follows. Firstly, an important invariant:
1061 * If the MVar is full, then the blocking queue contains only
1062 * threads blocked on putMVar, and if the MVar is empty then the
1063 * blocking queue contains only threads blocked on takeMVar.
1066 * MVar empty : then add ourselves to the blocking queue
1067 * MVar full : remove the value from the MVar, and
1068 * blocking queue empty : return
1069 * blocking queue non-empty : perform the first blocked putMVar
1070 * from the queue, and wake up the
1071 * thread (MVar is now full again)
1073 * putMVar is just the dual of the above algorithm.
1075 * How do we "perform a putMVar"? Well, we have to fiddle around with
1076 * the stack of the thread waiting to do the putMVar. See
1077 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1078 * the stack layout, and the PerformPut and PerformTake macros below.
1080 * It is important that a blocked take or put is woken up with the
1081 * take/put already performed, because otherwise there would be a
1082 * small window of vulnerability where the thread could receive an
1083 * exception and never perform its take or put, and we'd end up with a
1086 * -------------------------------------------------------------------------- */
1090 /* args: R1 = MVar closure */
1092 if (StgMVar_value(R1) == stg_END_TSO_QUEUE_closure) {
1104 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, stg_newMVarzh );
1106 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1107 SET_HDR(mvar,stg_MVAR_DIRTY_info,W_[CCCS]);
1108 // MVARs start dirty: generation 0 has no mutable list
1109 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1110 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1111 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1116 #define PerformTake(tso, value) \
1117 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1118 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1120 #define PerformPut(tso,lval) \
1121 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1122 lval = W_[StgTSO_sp(tso) - WDS(1)];
1126 W_ mvar, val, info, tso;
1128 /* args: R1 = MVar closure */
1131 #if defined(THREADED_RTS)
1132 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1134 info = GET_INFO(mvar);
1137 if (info == stg_MVAR_CLEAN_info) {
1138 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr") [];
1141 /* If the MVar is empty, put ourselves on its blocking queue,
1142 * and wait until we're woken up.
1144 if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1145 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1146 StgMVar_head(mvar) = CurrentTSO;
1148 foreign "C" setTSOLink(MyCapability() "ptr",
1149 StgMVar_tail(mvar) "ptr",
1152 StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1153 StgTSO_block_info(CurrentTSO) = mvar;
1154 // write barrier for throwTo(), which looks at block_info
1155 // if why_blocked==BlockedOnMVar.
1156 prim %write_barrier() [];
1157 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1158 StgMVar_tail(mvar) = CurrentTSO;
1161 jump stg_block_takemvar;
1164 /* we got the value... */
1165 val = StgMVar_value(mvar);
1167 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1169 /* There are putMVar(s) waiting...
1170 * wake up the first thread on the queue
1172 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1174 /* actually perform the putMVar for the thread that we just woke up */
1175 tso = StgMVar_head(mvar);
1176 PerformPut(tso,StgMVar_value(mvar));
1178 if (TO_W_(StgTSO_dirty(tso)) == 0) {
1179 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1182 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1183 StgMVar_head(mvar) "ptr", 1) [];
1184 StgMVar_head(mvar) = tso;
1186 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1187 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1190 #if defined(THREADED_RTS)
1191 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1193 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1199 /* No further putMVars, MVar is now empty */
1200 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1202 #if defined(THREADED_RTS)
1203 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1205 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1215 W_ mvar, val, info, tso;
1217 /* args: R1 = MVar closure */
1221 #if defined(THREADED_RTS)
1222 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1224 info = GET_INFO(mvar);
1227 if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1228 #if defined(THREADED_RTS)
1229 unlockClosure(mvar, info);
1231 /* HACK: we need a pointer to pass back,
1232 * so we abuse NO_FINALIZER_closure
1234 RET_NP(0, stg_NO_FINALIZER_closure);
1237 if (info == stg_MVAR_CLEAN_info) {
1238 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1241 /* we got the value... */
1242 val = StgMVar_value(mvar);
1244 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1246 /* There are putMVar(s) waiting...
1247 * wake up the first thread on the queue
1249 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1251 /* actually perform the putMVar for the thread that we just woke up */
1252 tso = StgMVar_head(mvar);
1253 PerformPut(tso,StgMVar_value(mvar));
1254 if (TO_W_(StgTSO_dirty(tso)) == 0) {
1255 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1258 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1259 StgMVar_head(mvar) "ptr", 1) [];
1260 StgMVar_head(mvar) = tso;
1262 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1263 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1265 #if defined(THREADED_RTS)
1266 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1268 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1273 /* No further putMVars, MVar is now empty */
1274 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1275 #if defined(THREADED_RTS)
1276 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1278 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1288 W_ mvar, val, info, tso;
1290 /* args: R1 = MVar, R2 = value */
1294 #if defined(THREADED_RTS)
1295 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1297 info = GET_INFO(mvar);
1300 if (info == stg_MVAR_CLEAN_info) {
1301 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1304 if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
1305 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1306 StgMVar_head(mvar) = CurrentTSO;
1308 foreign "C" setTSOLink(MyCapability() "ptr",
1309 StgMVar_tail(mvar) "ptr",
1312 StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1313 StgTSO_block_info(CurrentTSO) = mvar;
1314 // write barrier for throwTo(), which looks at block_info
1315 // if why_blocked==BlockedOnMVar.
1316 prim %write_barrier() [];
1317 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1318 StgMVar_tail(mvar) = CurrentTSO;
1322 jump stg_block_putmvar;
1325 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1327 /* There are takeMVar(s) waiting: wake up the first one
1329 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1331 /* actually perform the takeMVar */
1332 tso = StgMVar_head(mvar);
1333 PerformTake(tso, val);
1334 if (TO_W_(StgTSO_dirty(tso)) == 0) {
1335 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1338 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1339 StgMVar_head(mvar) "ptr", 1) [];
1340 StgMVar_head(mvar) = tso;
1342 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1343 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1346 #if defined(THREADED_RTS)
1347 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1349 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1351 jump %ENTRY_CODE(Sp(0));
1355 /* No further takes, the MVar is now full. */
1356 StgMVar_value(mvar) = val;
1358 #if defined(THREADED_RTS)
1359 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1361 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1363 jump %ENTRY_CODE(Sp(0));
1366 /* ToDo: yield afterward for better communication performance? */
1374 /* args: R1 = MVar, R2 = value */
1377 #if defined(THREADED_RTS)
1378 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [R2];
1380 info = GET_INFO(mvar);
1383 if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
1384 #if defined(THREADED_RTS)
1385 unlockClosure(mvar, info);
1390 if (info == stg_MVAR_CLEAN_info) {
1391 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1394 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1396 /* There are takeMVar(s) waiting: wake up the first one
1398 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1400 /* actually perform the takeMVar */
1401 tso = StgMVar_head(mvar);
1402 PerformTake(tso, R2);
1403 if (TO_W_(StgTSO_dirty(tso)) == 0) {
1404 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1407 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1408 StgMVar_head(mvar) "ptr", 1) [];
1409 StgMVar_head(mvar) = tso;
1411 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1412 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1415 #if defined(THREADED_RTS)
1416 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1418 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1423 /* No further takes, the MVar is now full. */
1424 StgMVar_value(mvar) = R2;
1426 #if defined(THREADED_RTS)
1427 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1429 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1434 /* ToDo: yield afterward for better communication performance? */
1438 /* -----------------------------------------------------------------------------
1439 Stable pointer primitives
1440 ------------------------------------------------------------------------- */
1442 stg_makeStableNamezh
1446 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, stg_makeStableNamezh );
1448 (index) = foreign "C" lookupStableName(R1 "ptr") [];
1450 /* Is there already a StableName for this heap object?
1451 * stable_ptr_table is a pointer to an array of snEntry structs.
1453 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1454 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1455 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1456 StgStableName_sn(sn_obj) = index;
1457 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1459 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1470 MAYBE_GC(R1_PTR, stg_makeStablePtrzh);
1471 ("ptr" sp) = foreign "C" getStablePtr(R1 "ptr") [];
1475 stg_deRefStablePtrzh
1477 /* Args: R1 = the stable ptr */
1480 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1484 /* -----------------------------------------------------------------------------
1485 Bytecode object primitives
1486 ------------------------------------------------------------------------- */
1496 W_ bco, bitmap_arr, bytes, words;
1500 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1503 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R5_PTR, stg_newBCOzh );
1505 bco = Hp - bytes + WDS(1);
1506 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1508 StgBCO_instrs(bco) = R1;
1509 StgBCO_literals(bco) = R2;
1510 StgBCO_ptrs(bco) = R3;
1511 StgBCO_arity(bco) = HALF_W_(R4);
1512 StgBCO_size(bco) = HALF_W_(words);
1514 // Copy the arity/bitmap info into the BCO
1518 if (i < StgArrWords_words(bitmap_arr)) {
1519 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1530 // R1 = the BCO# for the AP
1534 // This function is *only* used to wrap zero-arity BCOs in an
1535 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1536 // saturated and always points directly to a FUN or BCO.
1537 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1538 StgBCO_arity(R1) == HALF_W_(0));
1540 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, stg_mkApUpd0zh);
1541 TICK_ALLOC_UP_THK(0, 0);
1542 CCCS_ALLOC(SIZEOF_StgAP);
1544 ap = Hp - SIZEOF_StgAP + WDS(1);
1545 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1547 StgAP_n_args(ap) = HALF_W_(0);
1555 /* args: R1 = closure to analyze */
1556 // TODO: Consider the absence of ptrs or nonptrs as a special case ?
1558 W_ info, ptrs, nptrs, p, ptrs_arr, nptrs_arr;
1559 info = %GET_STD_INFO(UNTAG(R1));
1561 // Some closures have non-standard layout, so we omit those here.
1563 type = TO_W_(%INFO_TYPE(info));
1564 switch [0 .. N_CLOSURE_TYPES] type {
1565 case THUNK_SELECTOR : {
1570 case THUNK, THUNK_1_0, THUNK_0_1, THUNK_2_0, THUNK_1_1,
1571 THUNK_0_2, THUNK_STATIC, AP, PAP, AP_STACK, BCO : {
1577 ptrs = TO_W_(%INFO_PTRS(info));
1578 nptrs = TO_W_(%INFO_NPTRS(info));
1583 W_ ptrs_arr_sz, nptrs_arr_sz;
1584 nptrs_arr_sz = SIZEOF_StgArrWords + WDS(nptrs);
1585 ptrs_arr_sz = SIZEOF_StgMutArrPtrs + WDS(ptrs);
1587 ALLOC_PRIM (ptrs_arr_sz + nptrs_arr_sz, R1_PTR, stg_unpackClosurezh);
1592 ptrs_arr = Hp - nptrs_arr_sz - ptrs_arr_sz + WDS(1);
1593 nptrs_arr = Hp - nptrs_arr_sz + WDS(1);
1595 SET_HDR(ptrs_arr, stg_MUT_ARR_PTRS_FROZEN_info, W_[CCCS]);
1596 StgMutArrPtrs_ptrs(ptrs_arr) = ptrs;
1600 W_[ptrs_arr + SIZEOF_StgMutArrPtrs + WDS(p)] = StgClosure_payload(clos,p);
1605 SET_HDR(nptrs_arr, stg_ARR_WORDS_info, W_[CCCS]);
1606 StgArrWords_words(nptrs_arr) = nptrs;
1610 W_[BYTE_ARR_CTS(nptrs_arr) + WDS(p)] = StgClosure_payload(clos, p+ptrs);
1614 RET_NPP(info, ptrs_arr, nptrs_arr);
1617 /* -----------------------------------------------------------------------------
1618 Thread I/O blocking primitives
1619 -------------------------------------------------------------------------- */
1621 /* Add a thread to the end of the blocked queue. (C-- version of the C
1622 * macro in Schedule.h).
1624 #define APPEND_TO_BLOCKED_QUEUE(tso) \
1625 ASSERT(StgTSO__link(tso) == END_TSO_QUEUE); \
1626 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
1627 W_[blocked_queue_hd] = tso; \
1629 foreign "C" setTSOLink(MyCapability() "ptr", W_[blocked_queue_tl] "ptr", tso) []; \
1631 W_[blocked_queue_tl] = tso;
1637 foreign "C" barf("waitRead# on threaded RTS") never returns;
1640 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1641 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1642 StgTSO_block_info(CurrentTSO) = R1;
1643 // No locking - we're not going to use this interface in the
1644 // threaded RTS anyway.
1645 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1646 jump stg_block_noregs;
1654 foreign "C" barf("waitWrite# on threaded RTS") never returns;
1657 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1658 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1659 StgTSO_block_info(CurrentTSO) = R1;
1660 // No locking - we're not going to use this interface in the
1661 // threaded RTS anyway.
1662 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1663 jump stg_block_noregs;
1668 STRING(stg_delayzh_malloc_str, "stg_delayzh")
1671 #ifdef mingw32_HOST_OS
1679 foreign "C" barf("delay# on threaded RTS") never returns;
1682 /* args: R1 (microsecond delay amount) */
1683 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1684 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
1686 #ifdef mingw32_HOST_OS
1688 /* could probably allocate this on the heap instead */
1689 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1690 stg_delayzh_malloc_str);
1691 (reqID) = foreign "C" addDelayRequest(R1);
1692 StgAsyncIOResult_reqID(ares) = reqID;
1693 StgAsyncIOResult_len(ares) = 0;
1694 StgAsyncIOResult_errCode(ares) = 0;
1695 StgTSO_block_info(CurrentTSO) = ares;
1697 /* Having all async-blocked threads reside on the blocked_queue
1698 * simplifies matters, so change the status to OnDoProc put the
1699 * delayed thread on the blocked_queue.
1701 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1702 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1703 jump stg_block_async_void;
1709 (time) = foreign "C" getourtimeofday() [R1];
1710 divisor = TO_W_(RtsFlags_MiscFlags_tickInterval(RtsFlags));
1714 divisor = divisor * 1000;
1715 target = ((R1 + divisor - 1) / divisor) /* divide rounding up */
1716 + time + 1; /* Add 1 as getourtimeofday rounds down */
1717 StgTSO_block_info(CurrentTSO) = target;
1719 /* Insert the new thread in the sleeping queue. */
1721 t = W_[sleeping_queue];
1723 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
1725 t = StgTSO__link(t);
1729 StgTSO__link(CurrentTSO) = t;
1731 W_[sleeping_queue] = CurrentTSO;
1733 foreign "C" setTSOLink(MyCapability() "ptr", prev "ptr", CurrentTSO) [];
1735 jump stg_block_noregs;
1737 #endif /* !THREADED_RTS */
1741 #ifdef mingw32_HOST_OS
1742 STRING(stg_asyncReadzh_malloc_str, "stg_asyncReadzh")
1749 foreign "C" barf("asyncRead# on threaded RTS") never returns;
1752 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1753 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1754 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1756 /* could probably allocate this on the heap instead */
1757 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1758 stg_asyncReadzh_malloc_str)
1760 (reqID) = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") [];
1761 StgAsyncIOResult_reqID(ares) = reqID;
1762 StgAsyncIOResult_len(ares) = 0;
1763 StgAsyncIOResult_errCode(ares) = 0;
1764 StgTSO_block_info(CurrentTSO) = ares;
1765 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1766 jump stg_block_async;
1770 STRING(stg_asyncWritezh_malloc_str, "stg_asyncWritezh")
1777 foreign "C" barf("asyncWrite# on threaded RTS") never returns;
1780 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1781 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1782 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1784 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1785 stg_asyncWritezh_malloc_str)
1787 (reqID) = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") [];
1789 StgAsyncIOResult_reqID(ares) = reqID;
1790 StgAsyncIOResult_len(ares) = 0;
1791 StgAsyncIOResult_errCode(ares) = 0;
1792 StgTSO_block_info(CurrentTSO) = ares;
1793 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1794 jump stg_block_async;
1798 STRING(stg_asyncDoProczh_malloc_str, "stg_asyncDoProczh")
1805 foreign "C" barf("asyncDoProc# on threaded RTS") never returns;
1808 /* args: R1 = proc, R2 = param */
1809 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1810 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1812 /* could probably allocate this on the heap instead */
1813 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1814 stg_asyncDoProczh_malloc_str)
1816 (reqID) = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") [];
1817 StgAsyncIOResult_reqID(ares) = reqID;
1818 StgAsyncIOResult_len(ares) = 0;
1819 StgAsyncIOResult_errCode(ares) = 0;
1820 StgTSO_block_info(CurrentTSO) = ares;
1821 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1822 jump stg_block_async;
1827 // noDuplicate# tries to ensure that none of the thunks under
1828 // evaluation by the current thread are also under evaluation by
1829 // another thread. It relies on *both* threads doing noDuplicate#;
1830 // the second one will get blocked if they are duplicating some work.
1833 SAVE_THREAD_STATE();
1834 ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16);
1835 foreign "C" threadPaused (MyCapability() "ptr", CurrentTSO "ptr") [];
1837 if (StgTSO_what_next(CurrentTSO) == ThreadKilled::I16) {
1838 jump stg_threadFinished;
1840 LOAD_THREAD_STATE();
1841 ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16);
1842 jump %ENTRY_CODE(Sp(0));
1848 W_ ap_stack, offset, val, ok;
1850 /* args: R1 = AP_STACK, R2 = offset */
1854 if (%INFO_PTR(ap_stack) == stg_AP_STACK_info) {
1856 val = StgAP_STACK_payload(ap_stack,offset);
1864 /* -----------------------------------------------------------------------------
1866 -------------------------------------------------------------------------- */
1868 // Write the cost center stack of the first argument on stderr; return
1869 // the second. Possibly only makes sense for already evaluated
1876 ccs = StgHeader_ccs(UNTAG(R1));
1877 foreign "C" fprintCCS_stderr(ccs "ptr") [R2];
1888 #ifndef THREADED_RTS
1889 RET_NP(0,ghczmprim_GHCziBool_False_closure);
1891 (spark) = foreign "C" findSpark(MyCapability());
1895 RET_NP(0,ghczmprim_GHCziBool_False_closure);
1904 #if defined(TRACING) || defined(DEBUG)
1905 foreign "C" traceUserMsg(MyCapability() "ptr", msg "ptr") [];
1907 jump %ENTRY_CODE(Sp(0));