1 /* -----------------------------------------------------------------------------
2 * $Id: Storage.h,v 1.28 2001/02/09 13:09:16 simonmar Exp $
4 * (c) The GHC Team, 1998-1999
6 * External Storage Manger Interface
8 * ---------------------------------------------------------------------------*/
14 #include "BlockAlloc.h"
15 #include "StoragePriv.h"
17 /* -----------------------------------------------------------------------------
18 Initialisation / De-initialisation
19 -------------------------------------------------------------------------- */
21 extern void initStorage(void);
22 extern void exitStorage(void);
24 /* -----------------------------------------------------------------------------
27 StgPtr allocate(int n) Allocates a chunk of contiguous store
28 n words long, returning a pointer to
29 the first word. Always succeeds.
31 Don't forget to TICK_ALLOC_XXX(...)
32 after calling allocate, for the
33 benefit of the ticky-ticky profiler.
35 rtsBool doYouWantToGC(void) Returns True if the storage manager is
36 ready to perform a GC, False otherwise.
38 lnat allocated_bytes(void) Returns the number of bytes allocated
39 via allocate() since the last GC.
40 Used in the reoprting of statistics.
42 SMP: allocate and doYouWantToGC can be used from STG code, they are
43 surrounded by a mutex.
44 -------------------------------------------------------------------------- */
46 extern StgPtr allocate(nat n);
47 static inline rtsBool doYouWantToGC(void)
49 return (alloc_blocks >= alloc_blocks_lim);
51 extern lnat allocated_bytes(void);
53 /* -----------------------------------------------------------------------------
54 ExtendNursery(hp,hplim) When hplim is reached, try to grab
55 some more allocation space. Returns
56 False if the allocation space is
57 exhausted, and the application should
58 call GarbageCollect().
59 -------------------------------------------------------------------------- */
61 #define ExtendNursery(hp,hplim) \
62 (CurrentNursery->free = (P_)(hp)+1, \
63 CurrentNursery->link == NULL ? rtsFalse : \
64 (CurrentNursery = CurrentNursery->link, \
65 OpenNursery(hp,hplim), \
68 extern void PleaseStopAllocating(void);
70 /* -----------------------------------------------------------------------------
71 Performing Garbage Collection
73 GarbageCollect(get_roots) Performs a garbage collection.
74 'get_roots' is called to find all the
75 roots that the system knows about.
77 StgClosure Called by get_roots on each root.
78 MarkRoot(StgClosure *p) Returns the new location of the root.
79 -------------------------------------------------------------------------- */
81 extern void GarbageCollect(void (*get_roots)(void),rtsBool force_major_gc);
82 extern StgClosure *MarkRoot(StgClosure *p);
84 /* Temporary measure to ensure we retain all the dynamically-loaded CAFs */
86 extern void markCafs( void );
89 /* -----------------------------------------------------------------------------
90 Generational garbage collection support
92 recordMutable(StgPtr p) Informs the garbage collector that a
93 previously immutable object has
94 become (permanently) mutable. Used
95 by thawArray and similar.
97 updateWithIndirection(p1,p2) Updates the object at p1 with an
98 indirection pointing to p2. This is
99 normally called for objects in an old
100 generation (>0) when they are updated.
102 updateWithPermIndirection(p1,p2) As above but uses a permanent indir.
104 -------------------------------------------------------------------------- */
107 * Storage manager mutex
110 extern pthread_mutex_t sm_mutex;
113 /* ToDo: shouldn't recordMutable and recordOldToNewPtrs acquire some
114 * kind of lock in the SMP case?
117 recordMutable(StgMutClosure *p)
122 ASSERT(p->header.info == &stg_WHITEHOLE_info || closure_MUTABLE(p));
124 ASSERT(closure_MUTABLE(p));
128 if (bd->gen->no > 0) {
129 p->mut_link = bd->gen->mut_list;
130 bd->gen->mut_list = p;
135 recordOldToNewPtrs(StgMutClosure *p)
140 if (bd->gen->no > 0) {
141 p->mut_link = bd->gen->mut_once_list;
142 bd->gen->mut_once_list = p;
147 #define updateWithIndirection(info, p1, p2) \
151 bd = Bdescr((P_)p1); \
152 if (bd->gen->no == 0) { \
153 ((StgInd *)p1)->indirectee = p2; \
154 SET_INFO(p1,&stg_IND_info); \
155 TICK_UPD_NEW_IND(); \
157 ((StgIndOldGen *)p1)->indirectee = p2; \
158 if (info != &stg_BLACKHOLE_BQ_info) { \
159 ACQUIRE_LOCK(&sm_mutex); \
160 ((StgIndOldGen *)p1)->mut_link = bd->gen->mut_once_list; \
161 bd->gen->mut_once_list = (StgMutClosure *)p1; \
162 RELEASE_LOCK(&sm_mutex); \
164 SET_INFO(p1,&stg_IND_OLDGEN_info); \
165 TICK_UPD_OLD_IND(); \
170 /* In the DEBUG case, we also zero out the slop of the old closure,
171 * so that the sanity checker can tell where the next closure is.
173 #define updateWithIndirection(info, p1, p2) \
177 bd = Bdescr((P_)p1); \
178 if (bd->gen->no == 0) { \
179 ((StgInd *)p1)->indirectee = p2; \
180 SET_INFO(p1,&stg_IND_info); \
181 TICK_UPD_NEW_IND(); \
183 if (info != &stg_BLACKHOLE_BQ_info) { \
185 StgInfoTable *inf = get_itbl(p1); \
186 nat np = inf->layout.payload.ptrs, \
187 nw = inf->layout.payload.nptrs, i; \
188 if (inf->type != THUNK_SELECTOR) { \
189 for (i = np; i < np + nw; i++) { \
190 ((StgClosure *)p1)->payload[i] = 0; \
194 ACQUIRE_LOCK(&sm_mutex); \
195 ((StgIndOldGen *)p1)->mut_link = bd->gen->mut_once_list; \
196 bd->gen->mut_once_list = (StgMutClosure *)p1; \
197 RELEASE_LOCK(&sm_mutex); \
199 ((StgIndOldGen *)p1)->indirectee = p2; \
200 SET_INFO(p1,&stg_IND_OLDGEN_info); \
201 TICK_UPD_OLD_IND(); \
206 /* Static objects all live in the oldest generation
208 #define updateWithStaticIndirection(info, p1, p2) \
210 ASSERT( ((StgMutClosure*)p1)->mut_link == NULL ); \
212 ACQUIRE_LOCK(&sm_mutex); \
213 ((StgMutClosure *)p1)->mut_link = oldest_gen->mut_once_list; \
214 oldest_gen->mut_once_list = (StgMutClosure *)p1; \
215 RELEASE_LOCK(&sm_mutex); \
217 ((StgInd *)p1)->indirectee = p2; \
218 SET_INFO((StgInd *)p1, &stg_IND_STATIC_info); \
219 TICK_UPD_STATIC_IND(); \
222 #if defined(TICKY_TICKY) || defined(PROFILING)
224 updateWithPermIndirection(const StgInfoTable *info, StgClosure *p1, StgClosure *p2)
229 if (bd->gen->no == 0) {
230 ((StgInd *)p1)->indirectee = p2;
231 SET_INFO(p1,&stg_IND_PERM_info);
232 TICK_UPD_NEW_PERM_IND(p1);
234 ((StgIndOldGen *)p1)->indirectee = p2;
235 if (info != &stg_BLACKHOLE_BQ_info) {
236 ACQUIRE_LOCK(&sm_mutex);
237 ((StgIndOldGen *)p1)->mut_link = bd->gen->mut_once_list;
238 bd->gen->mut_once_list = (StgMutClosure *)p1;
239 RELEASE_LOCK(&sm_mutex);
241 SET_INFO(p1,&stg_IND_OLDGEN_PERM_info);
242 TICK_UPD_OLD_PERM_IND();
247 /* -----------------------------------------------------------------------------
248 The CAF table - used to let us revert CAFs
249 -------------------------------------------------------------------------- */
252 void printMutOnceList(generation *gen);
253 void printMutableList(generation *gen);
256 /* --------------------------------------------------------------------------
257 Address space layout macros
258 --------------------------------------------------------------------------
260 Here are the assumptions GHC makes about address space layout.
261 Broadly, it thinks there are three sections:
263 CODE Read-only. Contains code and read-only data (such as
267 DATA Read-write data. Contains static closures (and on some
268 architectures, info tables too)
270 HEAP Dynamically-allocated closures
272 USER None of the above. The only way USER things arise right
273 now is when GHCi allocates a constructor info table, which
274 it does by mallocing them.
276 Three macros identify these three areas:
277 IS_CODE(p), IS_DATA(p), HEAP_ALLOCED(p)
279 HEAP_ALLOCED is called FOR EVERY SINGLE CLOSURE during GC.
282 Implementation of HEAP_ALLOCED
283 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
284 Concerning HEAP, most of the time (certainly under [Static] and [GHCi],
285 we ensure that the heap is allocated above some fixed address HEAP_BASE
286 (defined in MBlock.h). In this case we set TEXT_BEFORE_HEAP, and we
287 get a nice fast test.
289 Sometimes we can't be quite sure. For example in Windows, we can't
290 fix where our heap address space comes from. In this case we un-set
291 TEXT_BEFORE_HEAP. That makes it more expensive to test whether a pointer
292 comes from the HEAP section, because we need to look at the allocator's
293 address maps (see HEAP_ALLOCED macro)
295 Implementation of CODE and DATA
296 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
297 Concerning CODE and DATA, there are three main regimes:
299 [Static] Totally The segments are contiguous, and laid out
300 statically linked exactly as above
302 [GHCi] Static, GHCi may load new modules, but it knows the
303 except for GHCi address map, so for any given address it can
304 still tell which section it belongs to
306 [DLL] OS-supported Chunks of CODE and DATA may be mixed in
307 dynamic loading the address space, and we can't tell how
310 For the [Static] case, we assume memory is laid out like this
311 (in order of increasing addresses)
315 TEXT_SECTION_END_MARKER (usually _etext)
317 DATA_SECTION_END_MARKER (usually _end)
322 For the [GHCi] case, we have to consult GHCi's dynamic linker's
323 address maps, which is done by macros
324 is_dynamically_loaded_code_or_rodata_ptr
325 is_dynamically_loaded_code_or_rwdata_ptr
327 For the [DLL] case, IS_CODE and IS_DATA are really not usable at all.
331 #undef TEXT_BEFORE_HEAP
332 #ifndef mingw32_TARGET_OS
333 #define TEXT_BEFORE_HEAP 1
336 extern void* TEXT_SECTION_END_MARKER_DECL;
337 extern void* DATA_SECTION_END_MARKER_DECL;
340 /* Take into account code sections in dynamically loaded object files. */
341 #define IS_CODE_PTR(p) ( ((P_)(p) < (P_)&TEXT_SECTION_END_MARKER) \
342 || is_dynamically_loaded_code_or_rodata_ptr((char *)p) )
343 #define IS_DATA_PTR(p) ( ((P_)(p) >= (P_)&TEXT_SECTION_END_MARKER && \
344 (P_)(p) < (P_)&DATA_SECTION_END_MARKER) \
345 || is_dynamically_loaded_rwdata_ptr((char *)p) )
346 #define IS_USER_PTR(p) ( ((P_)(p) >= (P_)&DATA_SECTION_END_MARKER) \
347 && is_not_dynamically_loaded_ptr((char *)p) )
349 #define IS_CODE_PTR(p) ((P_)(p) < (P_)&TEXT_SECTION_END_MARKER)
350 #define IS_DATA_PTR(p) ((P_)(p) >= (P_)&TEXT_SECTION_END_MARKER && (P_)(p) < (P_)&DATA_SECTION_END_MARKER)
351 #define IS_USER_PTR(p) ((P_)(p) >= (P_)&DATA_SECTION_END_MARKER)
354 /* The HEAP_ALLOCED test below is called FOR EVERY SINGLE CLOSURE
355 * during GC. It needs to be FAST.
357 * BEWARE: when we're dynamically loading code (for GHCi), make sure
358 * that we don't load any code above HEAP_BASE, or this test won't work.
360 #ifdef TEXT_BEFORE_HEAP
361 # define HEAP_ALLOCED(x) ((StgPtr)(x) >= (StgPtr)(HEAP_BASE))
363 extern int is_heap_alloced(const void* x);
364 # define HEAP_ALLOCED(x) (is_heap_alloced(x))
368 /* --------------------------------------------------------------------------
369 Macros for distinguishing data pointers from code pointers
370 --------------------------------------------------------------------------
374 The garbage collector needs to make some critical distinctions between pointers.
375 In particular we need
377 LOOKS_LIKE_GHC_INFO(p) p points to an info table
379 For both of these macros, p is
380 *either* a pointer to a closure (static or heap allocated)
381 *or* a return address on the (Haskell) stack
383 (Return addresses are in fact info-pointers, so that the Haskell stack
384 looks very like a chunk of heap.)
386 The garbage collector uses LOOKS_LIKE_GHC_INFO when walking the stack, as it
387 walks over the "pending arguments" on its way to the next return address.
388 It is called moderately often, but not as often as HEAP_ALLOCED
390 ToDo: LOOKS_LIKE_GHC_INFO(p) does not return True when p points to a
391 constructor info table allocated by GHCi. We should really rename
392 LOOKS_LIKE_GHC_INFO to LOOKS_LIKE_GHC_RETURN_INFO.
396 LOOKS_LIKE_GHC_INFO is more complicated because of the need to distinguish
397 between static closures and info tables. It's a known portability problem.
398 We have three approaches:
400 Plan A: Address-space partitioning.
401 Keep info tables in the (single, contiguous) text segment: IS_CODE_PTR(p)
402 and static closures in the (single, contiguous) data segment: IS_DATA_PTR(p)
404 Plan A can fail for two reasons:
405 * In many environments (eg. dynamic loading),
406 text and data aren't in a single contiguous range.
407 * When we compile through vanilla C (no mangling) we sometimes
408 can't guaranteee to put info tables in the text section. This
409 happens eg. on MacOS where the C compiler refuses to put const
410 data in the text section if it has any code pointers in it
411 (which info tables do *only* when we're compiling without
412 TABLES_NEXT_TO_CODE).
414 Hence, Plan B: (compile-via-C-with-mangling, or native code generation)
415 Put a zero word before each static closure.
416 When compiling to native code, or via C-with-mangling, info tables
417 are laid out "backwards" from the address specified in the info pointer
418 (the entry code goes forward from the info pointer). Hence, the word
419 before the one referenced the info pointer is part of the info table,
420 and is guaranteed non-zero.
422 For reasons nobody seems to fully understand, the statically-allocated tables
423 of INTLIKE and CHARLIKE closures can't have this zero word, so we
424 have to test separately for them.
426 Plan B fails altogether for the compile-through-vanilla-C route, because
427 info tables aren't laid out backwards.
430 Hence, Plan C: (unregisterised, compile-through-vanilla-C route only)
431 If we didn't manage to get info tables into the text section, then
432 we can distinguish between a static closure pointer and an info
433 pointer as follows: the first word of an info table is a code pointer,
434 and therefore in text space, whereas the first word of a closure pointer
435 is an info pointer, and therefore not. Shazam!
439 /* When working with Win32 DLLs, static closures are identified by
440 being prefixed with a zero word. This is needed so that we can
441 distinguish between pointers to static closures and (reversed!)
444 This 'scheme' breaks down for closure tables such as CHARLIKE,
445 so we catch these separately.
447 LOOKS_LIKE_STATIC_CLOSURE()
448 - discriminates between static closures and info tbls
449 (needed by LOOKS_LIKE_GHC_INFO() below - [Win32 DLLs only.])
451 - distinguishes between static and heap allocated data.
453 #if defined(ENABLE_WIN32_DLL_SUPPORT)
454 /* definitely do not enable for mingw DietHEP */
455 #define LOOKS_LIKE_STATIC(r) (!(HEAP_ALLOCED(r)))
457 /* Tiresome predicates needed to check for pointers into the closure tables */
458 #define IS_CHARLIKE_CLOSURE(p) \
459 ( (P_)(p) >= (P_)stg_CHARLIKE_closure && \
460 (char*)(p) <= ((char*)stg_CHARLIKE_closure + \
461 (MAX_CHARLIKE-MIN_CHARLIKE) * sizeof(StgIntCharlikeClosure)) )
462 #define IS_INTLIKE_CLOSURE(p) \
463 ( (P_)(p) >= (P_)stg_INTLIKE_closure && \
464 (char*)(p) <= ((char*)stg_INTLIKE_closure + \
465 (MAX_INTLIKE-MIN_INTLIKE) * sizeof(StgIntCharlikeClosure)) )
467 #define LOOKS_LIKE_STATIC_CLOSURE(r) (((*(((unsigned long *)(r))-1)) == 0) || IS_CHARLIKE_CLOSURE(r) || IS_INTLIKE_CLOSURE(r))
469 #define LOOKS_LIKE_STATIC(r) IS_DATA_PTR(r)
470 #define LOOKS_LIKE_STATIC_CLOSURE(r) IS_DATA_PTR(r)
474 /* -----------------------------------------------------------------------------
475 Macros for distinguishing infotables from closures.
477 You'd think it'd be easy to tell an info pointer from a closure pointer:
478 closures live on the heap and infotables are in read only memory. Right?
479 Wrong! Static closures live in read only memory and Hugs allocates
480 infotables for constructors on the (writable) C heap.
481 -------------------------------------------------------------------------- */
484 /* not accurate by any means, but stops the assertions failing... */
485 # define IS_HUGS_CONSTR_INFO(info) IS_USER_PTR(info)
487 # define IS_HUGS_CONSTR_INFO(info) 0 /* ToDo: more than mildly bogus */
490 /* LOOKS_LIKE_GHC_INFO is called moderately often during GC, but
491 * Certainly not as often as HEAP_ALLOCED.
493 #ifdef TEXT_BEFORE_HEAP /* needed for mingw DietHEP */
494 # define LOOKS_LIKE_GHC_INFO(info) IS_CODE_PTR(info)
496 # define LOOKS_LIKE_GHC_INFO(info) (!HEAP_ALLOCED(info) \
497 && !LOOKS_LIKE_STATIC_CLOSURE(info))
501 /* -----------------------------------------------------------------------------
502 Macros for calculating how big a closure will be (used during allocation)
503 -------------------------------------------------------------------------- */
505 /* ToDo: replace unsigned int by nat. The only fly in the ointment is that
506 * nat comes from Rts.h which many folk dont include. Sigh!
508 static __inline__ StgOffset AP_sizeW ( unsigned int n_args )
509 { return sizeofW(StgAP_UPD) + n_args; }
511 static __inline__ StgOffset PAP_sizeW ( unsigned int n_args )
512 { return sizeofW(StgPAP) + n_args; }
514 static __inline__ StgOffset CONSTR_sizeW( unsigned int p, unsigned int np )
515 { return sizeofW(StgHeader) + p + np; }
517 static __inline__ StgOffset THUNK_SELECTOR_sizeW ( void )
518 { return sizeofW(StgHeader) + MIN_UPD_SIZE; }
520 static __inline__ StgOffset BLACKHOLE_sizeW ( void )
521 { return sizeofW(StgHeader) + MIN_UPD_SIZE; }
523 /* --------------------------------------------------------------------------
525 * ------------------------------------------------------------------------*/
527 static __inline__ StgOffset sizeW_fromITBL( const StgInfoTable* itbl )
528 { return sizeofW(StgClosure)
529 + sizeofW(StgPtr) * itbl->layout.payload.ptrs
530 + sizeofW(StgWord) * itbl->layout.payload.nptrs; }
532 static __inline__ StgOffset pap_sizeW( StgPAP* x )
533 { return PAP_sizeW(x->n_args); }
535 static __inline__ StgOffset arr_words_sizeW( StgArrWords* x )
536 { return sizeofW(StgArrWords) + x->words; }
538 static __inline__ StgOffset mut_arr_ptrs_sizeW( StgMutArrPtrs* x )
539 { return sizeofW(StgMutArrPtrs) + x->ptrs; }
541 static __inline__ StgWord tso_sizeW ( StgTSO *tso )
542 { return TSO_STRUCT_SIZEW + tso->stack_size; }