2 * sgen-gc.c: Simple generational GC.
5 * Paolo Molaro (lupus@ximian.com)
6 * Rodrigo Kumpera (kumpera@gmail.com)
8 * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
9 * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
11 * Thread start/stop adapted from Boehm's GC:
12 * Copyright (c) 1994 by Xerox Corporation. All rights reserved.
13 * Copyright (c) 1996 by Silicon Graphics. All rights reserved.
14 * Copyright (c) 1998 by Fergus Henderson. All rights reserved.
15 * Copyright (c) 2000-2004 by Hewlett-Packard Company. All rights reserved.
16 * Copyright 2001-2003 Ximian, Inc
17 * Copyright 2003-2010 Novell, Inc.
18 * Copyright 2011 Xamarin, Inc.
19 * Copyright (C) 2012 Xamarin Inc
21 * Licensed under the MIT license. See LICENSE file in the project root for full license information.
23 * Important: allocation provides always zeroed memory, having to do
24 * a memset after allocation is deadly for performance.
25 * Memory usage at startup is currently as follows:
27 * 64 KB internal space
29 * We should provide a small memory config with half the sizes
31 * We currently try to make as few mono assumptions as possible:
32 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
34 * 2) gc descriptor is the second word in the vtable (first word in the class)
35 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
36 * 4) there is a function to get an object's size and the number of
37 * elements in an array.
38 * 5) we know the special way bounds are allocated for complex arrays
39 * 6) we know about proxies and how to treat them when domains are unloaded
41 * Always try to keep stack usage to a minimum: no recursive behaviour
42 * and no large stack allocs.
44 * General description.
45 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
46 * When the nursery is full we start a nursery collection: this is performed with a
48 * When the old generation is full we start a copying GC of the old generation as well:
49 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
50 * in the future. Maybe we'll even do both during the same collection like IMMIX.
52 * The things that complicate this description are:
53 * *) pinned objects: we can't move them so we need to keep track of them
54 * *) no precise info of the thread stacks and registers: we need to be able to
55 * quickly find the objects that may be referenced conservatively and pin them
56 * (this makes the first issues more important)
57 * *) large objects are too expensive to be dealt with using copying GC: we handle them
58 * with mark/sweep during major collections
59 * *) some objects need to not move even if they are small (interned strings, Type handles):
60 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
61 * PinnedChunks regions
67 *) we could have a function pointer in MonoClass to implement
68 customized write barriers for value types
70 *) investigate the stuff needed to advance a thread to a GC-safe
71 point (single-stepping, read from unmapped memory etc) and implement it.
72 This would enable us to inline allocations and write barriers, for example,
73 or at least parts of them, like the write barrier checks.
74 We may need this also for handling precise info on stacks, even simple things
75 as having uninitialized data on the stack and having to wait for the prolog
76 to zero it. Not an issue for the last frame that we scan conservatively.
77 We could always not trust the value in the slots anyway.
79 *) modify the jit to save info about references in stack locations:
80 this can be done just for locals as a start, so that at least
81 part of the stack is handled precisely.
83 *) test/fix endianess issues
85 *) Implement a card table as the write barrier instead of remembered
86 sets? Card tables are not easy to implement with our current
87 memory layout. We have several different kinds of major heap
88 objects: Small objects in regular blocks, small objects in pinned
89 chunks and LOS objects. If we just have a pointer we have no way
90 to tell which kind of object it points into, therefore we cannot
91 know where its card table is. The least we have to do to make
92 this happen is to get rid of write barriers for indirect stores.
95 *) Get rid of write barriers for indirect stores. We can do this by
96 telling the GC to wbarrier-register an object once we do an ldloca
97 or ldelema on it, and to unregister it once it's not used anymore
98 (it can only travel downwards on the stack). The problem with
99 unregistering is that it needs to happen eventually no matter
100 what, even if exceptions are thrown, the thread aborts, etc.
101 Rodrigo suggested that we could do only the registering part and
102 let the collector find out (pessimistically) when it's safe to
103 unregister, namely when the stack pointer of the thread that
104 registered the object is higher than it was when the registering
105 happened. This might make for a good first implementation to get
106 some data on performance.
108 *) Some sort of blacklist support? Blacklists is a concept from the
109 Boehm GC: if during a conservative scan we find pointers to an
110 area which we might use as heap, we mark that area as unusable, so
111 pointer retention by random pinning pointers is reduced.
113 *) experiment with max small object size (very small right now - 2kb,
114 because it's tied to the max freelist size)
116 *) add an option to mmap the whole heap in one chunk: it makes for many
117 simplifications in the checks (put the nursery at the top and just use a single
118 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
119 not flexible (too much of the address space may be used by default or we can't
120 increase the heap as needed) and we'd need a race-free mechanism to return memory
121 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
122 was written to, munmap is needed, but the following mmap may not find the same segment
125 *) memzero the major fragments after restarting the world and optionally a smaller
128 *) investigate having fragment zeroing threads
130 *) separate locks for finalization and other minor stuff to reduce
133 *) try a different copying order to improve memory locality
135 *) a thread abort after a store but before the write barrier will
136 prevent the write barrier from executing
138 *) specialized dynamically generated markers/copiers
140 *) Dynamically adjust TLAB size to the number of threads. If we have
141 too many threads that do allocation, we might need smaller TLABs,
142 and we might get better performance with larger TLABs if we only
143 have a handful of threads. We could sum up the space left in all
144 assigned TLABs and if that's more than some percentage of the
145 nursery size, reduce the TLAB size.
147 *) Explore placing unreachable objects on unused nursery memory.
148 Instead of memset'ng a region to zero, place an int[] covering it.
149 A good place to start is add_nursery_frag. The tricky thing here is
150 placing those objects atomically outside of a collection.
152 *) Allocation should use asymmetric Dekker synchronization:
153 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
154 This should help weak consistency archs.
161 #define _XOPEN_SOURCE
162 #define _DARWIN_C_SOURCE
168 #ifdef HAVE_PTHREAD_H
171 #ifdef HAVE_PTHREAD_NP_H
172 #include <pthread_np.h>
180 #include "mono/sgen/sgen-gc.h"
181 #include "mono/sgen/sgen-cardtable.h"
182 #include "mono/sgen/sgen-protocol.h"
183 #include "mono/sgen/sgen-memory-governor.h"
184 #include "mono/sgen/sgen-hash-table.h"
185 #include "mono/sgen/sgen-cardtable.h"
186 #include "mono/sgen/sgen-pinning.h"
187 #include "mono/sgen/sgen-workers.h"
188 #include "mono/sgen/sgen-client.h"
189 #include "mono/sgen/sgen-pointer-queue.h"
190 #include "mono/sgen/gc-internal-agnostic.h"
191 #include "mono/utils/mono-proclib.h"
192 #include "mono/utils/mono-memory-model.h"
193 #include "mono/utils/hazard-pointer.h"
195 #include <mono/utils/memcheck.h>
197 #undef pthread_create
199 #undef pthread_detach
202 * ######################################################################
203 * ######## Types and constants used by the GC.
204 * ######################################################################
207 /* 0 means not initialized, 1 is initialized, -1 means in progress */
208 static int gc_initialized = 0;
209 /* If set, check if we need to do something every X allocations */
210 gboolean has_per_allocation_action;
211 /* If set, do a heap check every X allocation */
212 guint32 verify_before_allocs = 0;
213 /* If set, do a minor collection before every X allocation */
214 guint32 collect_before_allocs = 0;
215 /* If set, do a whole heap check before each collection */
216 static gboolean whole_heap_check_before_collection = FALSE;
217 /* If set, do a remset consistency check at various opportunities */
218 static gboolean remset_consistency_checks = FALSE;
219 /* If set, do a mod union consistency check before each finishing collection pause */
220 static gboolean mod_union_consistency_check = FALSE;
221 /* If set, check whether mark bits are consistent after major collections */
222 static gboolean check_mark_bits_after_major_collection = FALSE;
223 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
224 static gboolean check_nursery_objects_pinned = FALSE;
225 /* If set, do a few checks when the concurrent collector is used */
226 static gboolean do_concurrent_checks = FALSE;
227 /* If set, do a plausibility check on the scan_starts before and after
229 static gboolean do_scan_starts_check = FALSE;
231 static gboolean disable_minor_collections = FALSE;
232 static gboolean disable_major_collections = FALSE;
233 static gboolean do_verify_nursery = FALSE;
234 static gboolean do_dump_nursery_content = FALSE;
235 static gboolean enable_nursery_canaries = FALSE;
237 static gboolean precleaning_enabled = TRUE;
239 #ifdef HEAVY_STATISTICS
240 guint64 stat_objects_alloced_degraded = 0;
241 guint64 stat_bytes_alloced_degraded = 0;
243 guint64 stat_copy_object_called_nursery = 0;
244 guint64 stat_objects_copied_nursery = 0;
245 guint64 stat_copy_object_called_major = 0;
246 guint64 stat_objects_copied_major = 0;
248 guint64 stat_scan_object_called_nursery = 0;
249 guint64 stat_scan_object_called_major = 0;
251 guint64 stat_slots_allocated_in_vain;
253 guint64 stat_nursery_copy_object_failed_from_space = 0;
254 guint64 stat_nursery_copy_object_failed_forwarded = 0;
255 guint64 stat_nursery_copy_object_failed_pinned = 0;
256 guint64 stat_nursery_copy_object_failed_to_space = 0;
258 static guint64 stat_wbarrier_add_to_global_remset = 0;
259 static guint64 stat_wbarrier_arrayref_copy = 0;
260 static guint64 stat_wbarrier_generic_store = 0;
261 static guint64 stat_wbarrier_generic_store_atomic = 0;
262 static guint64 stat_wbarrier_set_root = 0;
265 static guint64 stat_pinned_objects = 0;
267 static guint64 time_minor_pre_collection_fragment_clear = 0;
268 static guint64 time_minor_pinning = 0;
269 static guint64 time_minor_scan_remsets = 0;
270 static guint64 time_minor_scan_pinned = 0;
271 static guint64 time_minor_scan_roots = 0;
272 static guint64 time_minor_finish_gray_stack = 0;
273 static guint64 time_minor_fragment_creation = 0;
275 static guint64 time_major_pre_collection_fragment_clear = 0;
276 static guint64 time_major_pinning = 0;
277 static guint64 time_major_scan_pinned = 0;
278 static guint64 time_major_scan_roots = 0;
279 static guint64 time_major_scan_mod_union = 0;
280 static guint64 time_major_finish_gray_stack = 0;
281 static guint64 time_major_free_bigobjs = 0;
282 static guint64 time_major_los_sweep = 0;
283 static guint64 time_major_sweep = 0;
284 static guint64 time_major_fragment_creation = 0;
286 static guint64 time_max = 0;
288 static SGEN_TV_DECLARE (time_major_conc_collection_start);
289 static SGEN_TV_DECLARE (time_major_conc_collection_end);
291 int gc_debug_level = 0;
293 static char* gc_params_options;
294 static char* gc_debug_options;
298 mono_gc_flush_info (void)
300 fflush (gc_debug_file);
304 #define TV_DECLARE SGEN_TV_DECLARE
305 #define TV_GETTIME SGEN_TV_GETTIME
306 #define TV_ELAPSED SGEN_TV_ELAPSED
308 static SGEN_TV_DECLARE (sgen_init_timestamp);
310 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
312 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
313 #define object_is_pinned SGEN_OBJECT_IS_PINNED
314 #define pin_object SGEN_PIN_OBJECT
316 #define ptr_in_nursery sgen_ptr_in_nursery
318 #define LOAD_VTABLE SGEN_LOAD_VTABLE
321 nursery_canaries_enabled (void)
323 return enable_nursery_canaries;
326 #define safe_object_get_size sgen_safe_object_get_size
328 #if defined(HAVE_CONC_GC_AS_DEFAULT)
329 /* Use concurrent major on deskstop platforms */
330 #define DEFAULT_MAJOR_INIT sgen_marksweep_conc_init
331 #define DEFAULT_MAJOR_NAME "marksweep-conc"
333 #define DEFAULT_MAJOR_INIT sgen_marksweep_init
334 #define DEFAULT_MAJOR_NAME "marksweep"
338 * ######################################################################
339 * ######## Global data.
340 * ######################################################################
342 MonoCoopMutex gc_mutex;
344 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
346 size_t degraded_mode = 0;
348 static mword bytes_pinned_from_failed_allocation = 0;
350 GCMemSection *nursery_section = NULL;
351 static volatile mword lowest_heap_address = ~(mword)0;
352 static volatile mword highest_heap_address = 0;
354 MonoCoopMutex sgen_interruption_mutex;
356 int current_collection_generation = -1;
357 static volatile gboolean concurrent_collection_in_progress = FALSE;
359 /* objects that are ready to be finalized */
360 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
361 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
363 /* registered roots: the key to the hash is the root start address */
365 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
367 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
368 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
369 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
370 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
372 static mword roots_size = 0; /* amount of memory in the root set */
374 /* The size of a TLAB */
375 /* The bigger the value, the less often we have to go to the slow path to allocate a new
376 * one, but the more space is wasted by threads not allocating much memory.
378 * FIXME: Make this self-tuning for each thread.
380 guint32 tlab_size = (1024 * 4);
382 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
384 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
386 #define ALIGN_UP SGEN_ALIGN_UP
388 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
389 MonoNativeThreadId main_gc_thread = NULL;
392 /*Object was pinned during the current collection*/
393 static mword objects_pinned;
396 * ######################################################################
397 * ######## Macros and function declarations.
398 * ######################################################################
401 /* forward declarations */
402 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
404 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
405 static void finish_gray_stack (int generation, ScanCopyContext ctx);
408 SgenMajorCollector major_collector;
409 SgenMinorCollector sgen_minor_collector;
411 static SgenRememberedSet remset;
414 * The gray queue a worker job must use. If we're not parallel or
415 * concurrent, we use the main gray queue.
417 static SgenGrayQueue*
418 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
421 return &worker_data->private_gray_queue;
422 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
423 return default_gray_queue;
427 gray_queue_enable_redirect (SgenGrayQueue *queue)
429 SGEN_ASSERT (0, concurrent_collection_in_progress, "Where are we redirecting the gray queue to, without a concurrent collection?");
431 sgen_gray_queue_set_alloc_prepare (queue, sgen_workers_take_from_queue_and_awake);
432 sgen_workers_take_from_queue_and_awake (queue);
436 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
438 while (start < end) {
442 if (!*(void**)start) {
443 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
448 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
454 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
455 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
456 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
457 callback ((GCObject*)obj, size, data);
458 CANARIFY_SIZE (size);
460 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
468 * sgen_add_to_global_remset:
470 * The global remset contains locations which point into newspace after
471 * a minor collection. This can happen if the objects they point to are pinned.
473 * LOCKING: If called from a parallel collector, the global remset
474 * lock must be held. For serial collectors that is not necessary.
477 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
479 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
481 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
483 if (!major_collector.is_concurrent) {
484 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
486 if (current_collection_generation == -1)
487 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
490 if (!object_is_pinned (obj))
491 SGEN_ASSERT (5, sgen_minor_collector.is_split || sgen_concurrent_collection_in_progress (), "Non-pinned objects can only remain in nursery if it is a split nursery");
492 else if (sgen_cement_lookup_or_register (obj))
495 remset.record_pointer (ptr);
497 sgen_pin_stats_register_global_remset (obj);
499 SGEN_LOG (8, "Adding global remset for %p", ptr);
500 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
504 * sgen_drain_gray_stack:
506 * Scan objects in the gray stack until the stack is empty. This should be called
507 * frequently after each object is copied, to achieve better locality and cache
512 sgen_drain_gray_stack (ScanCopyContext ctx)
514 ScanObjectFunc scan_func = ctx.ops->scan_object;
515 SgenGrayQueue *queue = ctx.queue;
517 if (ctx.ops->drain_gray_stack)
518 return ctx.ops->drain_gray_stack (queue);
523 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
526 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
527 scan_func (obj, desc, queue);
533 * Addresses in the pin queue are already sorted. This function finds
534 * the object header for each address and pins the object. The
535 * addresses must be inside the nursery section. The (start of the)
536 * address array is overwritten with the addresses of the actually
537 * pinned objects. Return the number of pinned objects.
540 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
542 GCMemSection *section = nursery_section;
543 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
544 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
545 void *start_nursery = section->data;
546 void *end_nursery = section->next_data;
551 void *pinning_front = start_nursery;
553 void **definitely_pinned = start;
554 ScanObjectFunc scan_func = ctx.ops->scan_object;
555 SgenGrayQueue *queue = ctx.queue;
557 sgen_nursery_allocator_prepare_for_pinning ();
559 while (start < end) {
560 GCObject *obj_to_pin = NULL;
561 size_t obj_to_pin_size = 0;
566 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
567 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
574 SGEN_LOG (5, "Considering pinning addr %p", addr);
575 /* We've already processed everything up to pinning_front. */
576 if (addr < pinning_front) {
582 * Find the closest scan start <= addr. We might search backward in the
583 * scan_starts array because entries might be NULL. In the worst case we
584 * start at start_nursery.
586 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
587 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
588 search_start = (void*)section->scan_starts [idx];
589 if (!search_start || search_start > addr) {
592 search_start = section->scan_starts [idx];
593 if (search_start && search_start <= addr)
596 if (!search_start || search_start > addr)
597 search_start = start_nursery;
601 * If the pinning front is closer than the scan start we found, start
602 * searching at the front.
604 if (search_start < pinning_front)
605 search_start = pinning_front;
608 * Now addr should be in an object a short distance from search_start.
610 * search_start must point to zeroed mem or point to an object.
613 size_t obj_size, canarified_obj_size;
616 if (!*(void**)search_start) {
617 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
618 /* The loop condition makes sure we don't overrun addr. */
622 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
625 * Filler arrays are marked by an invalid sync word. We don't
626 * consider them for pinning. They are not delimited by canaries,
629 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
630 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
631 CANARIFY_SIZE (canarified_obj_size);
633 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
634 /* This is the object we're looking for. */
635 obj_to_pin = (GCObject*)search_start;
636 obj_to_pin_size = canarified_obj_size;
641 /* Skip to the next object */
642 search_start = (void*)((char*)search_start + canarified_obj_size);
643 } while (search_start <= addr);
645 /* We've searched past the address we were looking for. */
647 pinning_front = search_start;
648 goto next_pin_queue_entry;
652 * We've found an object to pin. It might still be a dummy array, but we
653 * can advance the pinning front in any case.
655 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
658 * If this is a dummy array marking the beginning of a nursery
659 * fragment, we don't pin it.
661 if (sgen_client_object_is_array_fill (obj_to_pin))
662 goto next_pin_queue_entry;
665 * Finally - pin the object!
667 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
668 if (do_scan_objects) {
669 scan_func (obj_to_pin, desc, queue);
671 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
672 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
673 binary_protocol_pin (obj_to_pin,
674 (gpointer)LOAD_VTABLE (obj_to_pin),
675 safe_object_get_size (obj_to_pin));
677 pin_object (obj_to_pin);
678 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
679 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
680 definitely_pinned [count] = obj_to_pin;
683 if (concurrent_collection_in_progress)
684 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
686 next_pin_queue_entry:
690 sgen_client_nursery_objects_pinned (definitely_pinned, count);
691 stat_pinned_objects += count;
696 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
700 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
703 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
704 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
708 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
709 * when we can't promote an object because we're out of memory.
712 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
714 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
717 * All pinned objects are assumed to have been staged, so we need to stage as well.
718 * Also, the count of staged objects shows that "late pinning" happened.
720 sgen_pin_stage_ptr (object);
722 SGEN_PIN_OBJECT (object);
723 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
726 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
728 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
731 /* Sort the addresses in array in increasing order.
732 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
735 sgen_sort_addresses (void **array, size_t size)
740 for (i = 1; i < size; ++i) {
743 size_t parent = (child - 1) / 2;
745 if (array [parent] >= array [child])
748 tmp = array [parent];
749 array [parent] = array [child];
756 for (i = size - 1; i > 0; --i) {
759 array [i] = array [0];
765 while (root * 2 + 1 <= end) {
766 size_t child = root * 2 + 1;
768 if (child < end && array [child] < array [child + 1])
770 if (array [root] >= array [child])
774 array [root] = array [child];
783 * Scan the memory between start and end and queue values which could be pointers
784 * to the area between start_nursery and end_nursery for later consideration.
785 * Typically used for thread stacks.
788 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
792 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
794 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
795 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
798 while (start < end) {
800 * *start can point to the middle of an object
801 * note: should we handle pointing at the end of an object?
802 * pinning in C# code disallows pointing at the end of an object
803 * but there is some small chance that an optimizing C compiler
804 * may keep the only reference to an object by pointing
805 * at the end of it. We ignore this small chance for now.
806 * Pointers to the end of an object are indistinguishable
807 * from pointers to the start of the next object in memory
808 * so if we allow that we'd need to pin two objects...
809 * We queue the pointer in an array, the
810 * array will then be sorted and uniqued. This way
811 * we can coalesce several pinning pointers and it should
812 * be faster since we'd do a memory scan with increasing
813 * addresses. Note: we can align the address to the allocation
814 * alignment, so the unique process is more effective.
816 mword addr = (mword)*start;
817 addr &= ~(ALLOC_ALIGN - 1);
818 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
819 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
820 sgen_pin_stage_ptr ((void*)addr);
821 binary_protocol_pin_stage (start, (void*)addr);
822 sgen_pin_stats_register_address ((char*)addr, pin_type);
828 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
832 * The first thing we do in a collection is to identify pinned objects.
833 * This function considers all the areas of memory that need to be
834 * conservatively scanned.
837 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
841 SGEN_LOG (2, "Scanning pinned roots (%d bytes, %d/%d entries)", (int)roots_size, roots_hash [ROOT_TYPE_NORMAL].num_entries, roots_hash [ROOT_TYPE_PINNED].num_entries);
842 /* objects pinned from the API are inside these roots */
843 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
844 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
845 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
846 } SGEN_HASH_TABLE_FOREACH_END;
847 /* now deal with the thread stacks
848 * in the future we should be able to conservatively scan only:
849 * *) the cpu registers
850 * *) the unmanaged stack frames
851 * *) the _last_ managed stack frame
852 * *) pointers slots in managed frames
854 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
858 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
860 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
861 ctx->ops->copy_or_mark_object (obj, ctx->queue);
865 * The memory area from start_root to end_root contains pointers to objects.
866 * Their position is precisely described by @desc (this means that the pointer
867 * can be either NULL or the pointer to the start of an object).
868 * This functions copies them to to_space updates them.
870 * This function is not thread-safe!
873 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
875 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
876 SgenGrayQueue *queue = ctx.queue;
878 switch (desc & ROOT_DESC_TYPE_MASK) {
879 case ROOT_DESC_BITMAP:
880 desc >>= ROOT_DESC_TYPE_SHIFT;
882 if ((desc & 1) && *start_root) {
883 copy_func ((GCObject**)start_root, queue);
884 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
890 case ROOT_DESC_COMPLEX: {
891 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
892 gsize bwords = (*bitmap_data) - 1;
893 void **start_run = start_root;
895 while (bwords-- > 0) {
896 gsize bmap = *bitmap_data++;
897 void **objptr = start_run;
899 if ((bmap & 1) && *objptr) {
900 copy_func ((GCObject**)objptr, queue);
901 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
906 start_run += GC_BITS_PER_WORD;
910 case ROOT_DESC_USER: {
911 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
912 marker (start_root, single_arg_user_copy_or_mark, &ctx);
915 case ROOT_DESC_RUN_LEN:
916 g_assert_not_reached ();
918 g_assert_not_reached ();
923 reset_heap_boundaries (void)
925 lowest_heap_address = ~(mword)0;
926 highest_heap_address = 0;
930 sgen_update_heap_boundaries (mword low, mword high)
935 old = lowest_heap_address;
938 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
941 old = highest_heap_address;
944 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
948 * Allocate and setup the data structures needed to be able to allocate objects
949 * in the nursery. The nursery is stored in nursery_section.
954 GCMemSection *section;
961 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
962 /* later we will alloc a larger area for the nursery but only activate
963 * what we need. The rest will be used as expansion if we have too many pinned
964 * objects in the existing nursery.
966 /* FIXME: handle OOM */
967 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
969 alloc_size = sgen_nursery_size;
971 /* If there isn't enough space even for the nursery we should simply abort. */
972 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
974 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
975 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
976 SGEN_LOG (4, "Expanding nursery size (%p-%p): %lu, total: %lu", data, data + alloc_size, (unsigned long)sgen_nursery_size, (unsigned long)sgen_gc_get_total_heap_allocation ());
977 section->data = section->next_data = data;
978 section->size = alloc_size;
979 section->end_data = data + sgen_nursery_size;
980 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
981 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
982 section->num_scan_start = scan_starts;
984 nursery_section = section;
986 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
990 mono_gc_get_logfile (void)
992 return gc_debug_file;
996 mono_gc_params_set (const char* options)
998 if (gc_params_options)
999 g_free (gc_params_options);
1001 gc_params_options = g_strdup (options);
1005 mono_gc_debug_set (const char* options)
1007 if (gc_debug_options)
1008 g_free (gc_debug_options);
1010 gc_debug_options = g_strdup (options);
1014 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1016 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1017 SgenGrayQueue *queue = ctx.queue;
1020 for (i = 0; i < fin_queue->next_slot; ++i) {
1021 GCObject *obj = (GCObject *)fin_queue->data [i];
1024 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1025 copy_func ((GCObject**)&fin_queue->data [i], queue);
1030 generation_name (int generation)
1032 switch (generation) {
1033 case GENERATION_NURSERY: return "nursery";
1034 case GENERATION_OLD: return "old";
1035 default: g_assert_not_reached ();
1040 sgen_generation_name (int generation)
1042 return generation_name (generation);
1046 finish_gray_stack (int generation, ScanCopyContext ctx)
1050 int done_with_ephemerons, ephemeron_rounds = 0;
1051 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1052 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1053 SgenGrayQueue *queue = ctx.queue;
1055 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1057 * We copied all the reachable objects. Now it's the time to copy
1058 * the objects that were not referenced by the roots, but by the copied objects.
1059 * we built a stack of objects pointed to by gray_start: they are
1060 * additional roots and we may add more items as we go.
1061 * We loop until gray_start == gray_objects which means no more objects have
1062 * been added. Note this is iterative: no recursion is involved.
1063 * We need to walk the LO list as well in search of marked big objects
1064 * (use a flag since this is needed only on major collections). We need to loop
1065 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1066 * To achieve better cache locality and cache usage, we drain the gray stack
1067 * frequently, after each object is copied, and just finish the work here.
1069 sgen_drain_gray_stack (ctx);
1071 SGEN_LOG (2, "%s generation done", generation_name (generation));
1074 Reset bridge data, we might have lingering data from a previous collection if this is a major
1075 collection trigged by minor overflow.
1077 We must reset the gathered bridges since their original block might be evacuated due to major
1078 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1080 if (sgen_client_bridge_need_processing ())
1081 sgen_client_bridge_reset_data ();
1084 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1085 * to ensure they see the full set of live objects.
1087 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1090 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1091 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1092 * objects that are in fact reachable.
1094 done_with_ephemerons = 0;
1096 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1097 sgen_drain_gray_stack (ctx);
1099 } while (!done_with_ephemerons);
1101 if (sgen_client_bridge_need_processing ()) {
1102 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1103 sgen_drain_gray_stack (ctx);
1104 sgen_collect_bridge_objects (generation, ctx);
1105 if (generation == GENERATION_OLD)
1106 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1109 Do the first bridge step here, as the collector liveness state will become useless after that.
1111 An important optimization is to only proccess the possibly dead part of the object graph and skip
1112 over all live objects as we transitively know everything they point must be alive too.
1114 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1116 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1117 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1120 sgen_client_bridge_processing_stw_step ();
1124 Make sure we drain the gray stack before processing disappearing links and finalizers.
1125 If we don't make sure it is empty we might wrongly see a live object as dead.
1127 sgen_drain_gray_stack (ctx);
1130 We must clear weak links that don't track resurrection before processing object ready for
1131 finalization so they can be cleared before that.
1133 sgen_null_link_in_range (generation, ctx, FALSE);
1134 if (generation == GENERATION_OLD)
1135 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1138 /* walk the finalization queue and move also the objects that need to be
1139 * finalized: use the finalized objects as new roots so the objects they depend
1140 * on are also not reclaimed. As with the roots above, only objects in the nursery
1141 * are marked/copied.
1143 sgen_finalize_in_range (generation, ctx);
1144 if (generation == GENERATION_OLD)
1145 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1146 /* drain the new stack that might have been created */
1147 SGEN_LOG (6, "Precise scan of gray area post fin");
1148 sgen_drain_gray_stack (ctx);
1151 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1153 done_with_ephemerons = 0;
1155 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1156 sgen_drain_gray_stack (ctx);
1158 } while (!done_with_ephemerons);
1160 sgen_client_clear_unreachable_ephemerons (ctx);
1163 * We clear togglerefs only after all possible chances of revival are done.
1164 * This is semantically more inline with what users expect and it allows for
1165 * user finalizers to correctly interact with TR objects.
1167 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1170 SGEN_LOG (2, "Finalize queue handling scan for %s generation: %lld usecs %d ephemeron rounds", generation_name (generation), (long long)TV_ELAPSED (atv, btv), ephemeron_rounds);
1173 * handle disappearing links
1174 * Note we do this after checking the finalization queue because if an object
1175 * survives (at least long enough to be finalized) we don't clear the link.
1176 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1177 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1180 g_assert (sgen_gray_object_queue_is_empty (queue));
1182 sgen_null_link_in_range (generation, ctx, TRUE);
1183 if (generation == GENERATION_OLD)
1184 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1185 if (sgen_gray_object_queue_is_empty (queue))
1187 sgen_drain_gray_stack (ctx);
1190 g_assert (sgen_gray_object_queue_is_empty (queue));
1192 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1196 sgen_check_section_scan_starts (GCMemSection *section)
1199 for (i = 0; i < section->num_scan_start; ++i) {
1200 if (section->scan_starts [i]) {
1201 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1202 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1208 check_scan_starts (void)
1210 if (!do_scan_starts_check)
1212 sgen_check_section_scan_starts (nursery_section);
1213 major_collector.check_scan_starts ();
1217 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1221 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1222 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1223 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1224 } SGEN_HASH_TABLE_FOREACH_END;
1230 static gboolean inited = FALSE;
1235 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1237 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1238 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1239 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1240 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1241 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1242 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1244 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1245 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1246 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1247 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1248 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1249 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1250 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1251 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1252 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1253 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1255 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1257 #ifdef HEAVY_STATISTICS
1258 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1259 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1260 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1261 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1262 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1264 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1265 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1267 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1268 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1269 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1270 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1272 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1273 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1275 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1277 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1278 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1279 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1280 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1282 sgen_nursery_allocator_init_heavy_stats ();
1290 reset_pinned_from_failed_allocation (void)
1292 bytes_pinned_from_failed_allocation = 0;
1296 sgen_set_pinned_from_failed_allocation (mword objsize)
1298 bytes_pinned_from_failed_allocation += objsize;
1302 sgen_collection_is_concurrent (void)
1304 switch (current_collection_generation) {
1305 case GENERATION_NURSERY:
1307 case GENERATION_OLD:
1308 return concurrent_collection_in_progress;
1310 g_error ("Invalid current generation %d", current_collection_generation);
1316 sgen_concurrent_collection_in_progress (void)
1318 return concurrent_collection_in_progress;
1322 SgenThreadPoolJob job;
1323 SgenObjectOperations *ops;
1324 SgenGrayQueue *gc_thread_gray_queue;
1327 static ScanCopyContext
1328 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1330 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1332 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1336 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1338 remset.scan_remsets (scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job));
1346 } ScanFromRegisteredRootsJob;
1349 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1351 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1352 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1354 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1361 } ScanThreadDataJob;
1364 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1366 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1367 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1369 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1374 SgenPointerQueue *queue;
1375 } ScanFinalizerEntriesJob;
1378 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1380 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1381 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1383 scan_finalizer_entries (job_data->queue, ctx);
1387 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1389 ScanJob *job_data = (ScanJob*)job;
1390 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1392 g_assert (concurrent_collection_in_progress);
1393 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1397 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1399 ScanJob *job_data = (ScanJob*)job;
1400 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1402 g_assert (concurrent_collection_in_progress);
1403 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1407 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1409 ScanJob *job_data = (ScanJob*)job;
1410 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1412 g_assert (concurrent_collection_in_progress);
1414 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1415 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1417 sgen_scan_pin_queue_objects (ctx);
1421 workers_finish_callback (void)
1424 /* Mod union preclean job */
1425 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1426 sj->ops = sgen_workers_get_idle_func_object_ops ();
1427 sj->gc_thread_gray_queue = NULL;
1429 sgen_workers_enqueue_job (&sj->job, TRUE);
1433 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue, gboolean use_workers)
1436 sgen_workers_init_distribute_gray_queue ();
1437 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1441 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1443 ScanFromRegisteredRootsJob *scrrj;
1444 ScanThreadDataJob *stdj;
1445 ScanFinalizerEntriesJob *sfej;
1447 /* registered roots, this includes static fields */
1449 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1450 scrrj->scan_job.ops = ops;
1451 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1452 scrrj->heap_start = heap_start;
1453 scrrj->heap_end = heap_end;
1454 scrrj->root_type = ROOT_TYPE_NORMAL;
1455 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1457 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1458 scrrj->scan_job.ops = ops;
1459 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1460 scrrj->heap_start = heap_start;
1461 scrrj->heap_end = heap_end;
1462 scrrj->root_type = ROOT_TYPE_WBARRIER;
1463 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1467 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1468 stdj->scan_job.ops = ops;
1469 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1470 stdj->heap_start = heap_start;
1471 stdj->heap_end = heap_end;
1472 sgen_workers_enqueue_job (&stdj->scan_job.job, enqueue);
1474 /* Scan the list of objects ready for finalization. */
1476 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1477 sfej->scan_job.ops = ops;
1478 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1479 sfej->queue = &fin_ready_queue;
1480 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1482 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1483 sfej->scan_job.ops = ops;
1484 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1485 sfej->queue = &critical_fin_queue;
1486 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1490 * Perform a nursery collection.
1492 * Return whether any objects were late-pinned due to being out of memory.
1495 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue)
1497 gboolean needs_major;
1498 size_t max_garbage_amount;
1500 mword fragment_total;
1502 SgenGrayQueue gc_thread_gray_queue;
1503 SgenObjectOperations *object_ops;
1504 ScanCopyContext ctx;
1507 SGEN_TV_DECLARE (last_minor_collection_start_tv);
1508 SGEN_TV_DECLARE (last_minor_collection_end_tv);
1510 if (disable_minor_collections)
1513 TV_GETTIME (last_minor_collection_start_tv);
1514 atv = last_minor_collection_start_tv;
1516 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1518 if (sgen_concurrent_collection_in_progress ())
1519 object_ops = &sgen_minor_collector.serial_ops_with_concurrent_major;
1521 object_ops = &sgen_minor_collector.serial_ops;
1523 if (do_verify_nursery || do_dump_nursery_content)
1524 sgen_debug_verify_nursery (do_dump_nursery_content);
1526 current_collection_generation = GENERATION_NURSERY;
1528 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1530 reset_pinned_from_failed_allocation ();
1532 check_scan_starts ();
1534 sgen_nursery_alloc_prepare_for_minor ();
1538 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1539 /* FIXME: optimize later to use the higher address where an object can be present */
1540 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1542 SGEN_LOG (1, "Start nursery collection %d %p-%p, size: %d", gc_stats.minor_gc_count, sgen_get_nursery_start (), nursery_next, (int)(nursery_next - sgen_get_nursery_start ()));
1543 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1544 g_assert (nursery_section->size >= max_garbage_amount);
1546 /* world must be stopped already */
1548 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1550 sgen_client_pre_collection_checks ();
1552 nursery_section->next_data = nursery_next;
1554 major_collector.start_nursery_collection ();
1556 sgen_memgov_minor_collection_start ();
1558 init_gray_queue (&gc_thread_gray_queue, FALSE);
1559 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gc_thread_gray_queue);
1561 gc_stats.minor_gc_count ++;
1563 sgen_process_fin_stage_entries ();
1565 /* pin from pinned handles */
1566 sgen_init_pinning ();
1567 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1568 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1569 /* pin cemented objects */
1570 sgen_pin_cemented_objects ();
1571 /* identify pinned objects */
1572 sgen_optimize_pin_queue ();
1573 sgen_pinning_setup_section (nursery_section);
1575 pin_objects_in_nursery (FALSE, ctx);
1576 sgen_pinning_trim_queue_to_section (nursery_section);
1578 if (remset_consistency_checks)
1579 sgen_check_remset_consistency ();
1581 if (whole_heap_check_before_collection) {
1582 sgen_clear_nursery_fragments ();
1583 sgen_check_whole_heap (FALSE);
1587 time_minor_pinning += TV_ELAPSED (btv, atv);
1588 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1589 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1591 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1592 sj->ops = object_ops;
1593 sj->gc_thread_gray_queue = &gc_thread_gray_queue;
1594 sgen_workers_enqueue_job (&sj->job, FALSE);
1596 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1598 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1599 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1601 sgen_pin_stats_report ();
1603 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1604 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1607 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1609 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1612 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1614 finish_gray_stack (GENERATION_NURSERY, ctx);
1617 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1618 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1620 if (objects_pinned) {
1621 sgen_optimize_pin_queue ();
1622 sgen_pinning_setup_section (nursery_section);
1626 * This is the latest point at which we can do this check, because
1627 * sgen_build_nursery_fragments() unpins nursery objects again.
1629 if (remset_consistency_checks)
1630 sgen_check_remset_consistency ();
1632 /* walk the pin_queue, build up the fragment list of free memory, unmark
1633 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1636 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1637 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1638 if (!fragment_total)
1641 /* Clear TLABs for all threads */
1642 sgen_clear_tlabs ();
1644 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1646 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1647 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1649 if (remset_consistency_checks)
1650 sgen_check_major_refs ();
1652 major_collector.finish_nursery_collection ();
1654 TV_GETTIME (last_minor_collection_end_tv);
1655 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1657 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1659 /* prepare the pin queue for the next collection */
1660 sgen_finish_pinning ();
1661 if (sgen_have_pending_finalizers ()) {
1662 SGEN_LOG (4, "Finalizer-thread wakeup");
1663 sgen_client_finalize_notify ();
1665 sgen_pin_stats_reset ();
1666 /* clear cemented hash */
1667 sgen_cement_clear_below_threshold ();
1669 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1671 remset.finish_minor_collection ();
1673 check_scan_starts ();
1675 binary_protocol_flush_buffers (FALSE);
1677 sgen_memgov_minor_collection_end (reason, is_overflow);
1679 /*objects are late pinned because of lack of memory, so a major is a good call*/
1680 needs_major = objects_pinned > 0;
1681 current_collection_generation = -1;
1684 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1686 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1687 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1693 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1694 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1695 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1696 } CopyOrMarkFromRootsMode;
1699 major_copy_or_mark_from_roots (SgenGrayQueue *gc_thread_gray_queue, size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1704 /* FIXME: only use these values for the precise scan
1705 * note that to_space pointers should be excluded anyway...
1707 char *heap_start = NULL;
1708 char *heap_end = (char*)-1;
1709 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, gc_thread_gray_queue);
1710 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1712 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1714 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1715 /*This cleans up unused fragments */
1716 sgen_nursery_allocator_prepare_for_pinning ();
1718 if (do_concurrent_checks)
1719 sgen_debug_check_nursery_is_clean ();
1721 /* The concurrent collector doesn't touch the nursery. */
1722 sgen_nursery_alloc_prepare_for_major ();
1727 /* Pinning depends on this */
1728 sgen_clear_nursery_fragments ();
1730 if (whole_heap_check_before_collection)
1731 sgen_check_whole_heap (TRUE);
1734 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1736 if (!sgen_collection_is_concurrent ())
1737 nursery_section->next_data = sgen_get_nursery_end ();
1738 /* we should also coalesce scanning from sections close to each other
1739 * and deal with pointers outside of the sections later.
1744 sgen_client_pre_collection_checks ();
1746 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1747 /* Remsets are not useful for a major collection */
1748 remset.clear_cards ();
1751 sgen_process_fin_stage_entries ();
1754 sgen_init_pinning ();
1755 SGEN_LOG (6, "Collecting pinned addresses");
1756 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1757 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1758 /* Pin cemented objects that were forced */
1759 sgen_pin_cemented_objects ();
1761 sgen_optimize_pin_queue ();
1762 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1764 * Cemented objects that are in the pinned list will be marked. When
1765 * marking concurrently we won't mark mod-union cards for these objects.
1766 * Instead they will remain cemented until the next major collection,
1767 * when we will recheck if they are still pinned in the roots.
1769 sgen_cement_force_pinned ();
1772 sgen_client_collecting_major_1 ();
1775 * pin_queue now contains all candidate pointers, sorted and
1776 * uniqued. We must do two passes now to figure out which
1777 * objects are pinned.
1779 * The first is to find within the pin_queue the area for each
1780 * section. This requires that the pin_queue be sorted. We
1781 * also process the LOS objects and pinned chunks here.
1783 * The second, destructive, pass is to reduce the section
1784 * areas to pointers to the actually pinned objects.
1786 SGEN_LOG (6, "Pinning from sections");
1787 /* first pass for the sections */
1788 sgen_find_section_pin_queue_start_end (nursery_section);
1789 /* identify possible pointers to the insize of large objects */
1790 SGEN_LOG (6, "Pinning from large objects");
1791 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1793 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1794 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1796 if (sgen_los_object_is_pinned (bigobj->data)) {
1797 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1800 sgen_los_pin_object (bigobj->data);
1801 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1802 GRAY_OBJECT_ENQUEUE (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1803 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1804 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1805 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1806 (unsigned long)sgen_los_object_size (bigobj));
1808 sgen_client_pinned_los_object (bigobj->data);
1812 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1813 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1814 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1816 major_collector.pin_objects (gc_thread_gray_queue);
1817 if (old_next_pin_slot)
1818 *old_next_pin_slot = sgen_get_pinned_count ();
1821 time_major_pinning += TV_ELAPSED (atv, btv);
1822 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1823 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1825 major_collector.init_to_space ();
1827 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1828 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1829 if (sgen_workers_have_idle_work ()) {
1831 * We force the finish of the worker with the new object ops context
1832 * which can also do copying. We need to have finished pinning.
1834 /* FIXME Implement parallel copying and get rid of this ineffective hack */
1835 if (major_collector.is_parallel)
1836 sgen_workers_start_all_workers (&major_collector.major_ops_conc_par_start, NULL);
1838 sgen_workers_start_all_workers (object_ops, NULL);
1840 sgen_workers_join ();
1844 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1845 main_gc_thread = mono_native_thread_self ();
1848 sgen_client_collecting_major_2 ();
1851 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1853 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1855 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops, FALSE);
1858 time_major_scan_roots += TV_ELAPSED (atv, btv);
1861 * We start the concurrent worker after pinning and after we scanned the roots
1862 * in order to make sure that the worker does not finish before handling all
1865 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1866 if (precleaning_enabled) {
1867 sgen_workers_start_all_workers (object_ops, workers_finish_callback);
1869 sgen_workers_start_all_workers (object_ops, NULL);
1871 gray_queue_enable_redirect (gc_thread_gray_queue);
1874 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1877 /* Mod union card table */
1878 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1879 sj->ops = object_ops;
1880 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1881 sgen_workers_enqueue_job (&sj->job, FALSE);
1883 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1884 sj->ops = object_ops;
1885 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1886 sgen_workers_enqueue_job (&sj->job, FALSE);
1889 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1892 sgen_pin_stats_report ();
1894 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1895 sgen_finish_pinning ();
1897 sgen_pin_stats_reset ();
1899 if (do_concurrent_checks)
1900 sgen_debug_check_nursery_is_clean ();
1905 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
1907 SgenObjectOperations *object_ops;
1909 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1911 current_collection_generation = GENERATION_OLD;
1913 sgen_workers_assert_gray_queue_is_empty ();
1916 sgen_cement_reset ();
1919 g_assert (major_collector.is_concurrent);
1920 concurrent_collection_in_progress = TRUE;
1922 if (major_collector.is_parallel)
1923 object_ops = &major_collector.major_ops_conc_par_start;
1925 object_ops = &major_collector.major_ops_concurrent_start;
1928 object_ops = &major_collector.major_ops_serial;
1931 reset_pinned_from_failed_allocation ();
1933 sgen_memgov_major_collection_start (concurrent, reason);
1935 //count_ref_nonref_objs ();
1936 //consistency_check ();
1938 check_scan_starts ();
1941 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1942 gc_stats.major_gc_count ++;
1944 if (major_collector.start_major_collection)
1945 major_collector.start_major_collection ();
1947 major_copy_or_mark_from_roots (gc_thread_gray_queue, old_next_pin_slot, concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL, object_ops);
1951 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
1953 ScannedObjectCounts counts;
1954 SgenObjectOperations *object_ops;
1955 mword fragment_total;
1961 if (concurrent_collection_in_progress) {
1962 if (major_collector.is_parallel)
1963 object_ops = &major_collector.major_ops_conc_par_finish;
1965 object_ops = &major_collector.major_ops_concurrent_finish;
1967 major_copy_or_mark_from_roots (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1969 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1970 main_gc_thread = NULL;
1973 object_ops = &major_collector.major_ops_serial;
1976 sgen_workers_assert_gray_queue_is_empty ();
1978 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, gc_thread_gray_queue));
1980 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1982 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1984 if (objects_pinned) {
1985 g_assert (!concurrent_collection_in_progress);
1988 * This is slow, but we just OOM'd.
1990 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1991 * queue is laid out at this point.
1993 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1995 * We need to reestablish all pinned nursery objects in the pin queue
1996 * because they're needed for fragment creation. Unpinning happens by
1997 * walking the whole queue, so it's not necessary to reestablish where major
1998 * heap block pins are - all we care is that they're still in there
2001 sgen_optimize_pin_queue ();
2002 sgen_find_section_pin_queue_start_end (nursery_section);
2006 reset_heap_boundaries ();
2007 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
2009 /* walk the pin_queue, build up the fragment list of free memory, unmark
2010 * pinned objects as we go, memzero() the empty fragments so they are ready for the
2013 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
2014 if (!fragment_total)
2016 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
2018 if (do_concurrent_checks && concurrent_collection_in_progress)
2019 sgen_debug_check_nursery_is_clean ();
2021 /* prepare the pin queue for the next collection */
2022 sgen_finish_pinning ();
2024 /* Clear TLABs for all threads */
2025 sgen_clear_tlabs ();
2027 sgen_pin_stats_reset ();
2029 sgen_cement_clear_below_threshold ();
2031 if (check_mark_bits_after_major_collection)
2032 sgen_check_heap_marked (concurrent_collection_in_progress);
2035 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2037 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2038 sgen_memgov_major_pre_sweep ();
2041 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2046 time_major_los_sweep += TV_ELAPSED (atv, btv);
2048 major_collector.sweep ();
2050 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2053 time_major_sweep += TV_ELAPSED (btv, atv);
2055 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2057 if (sgen_have_pending_finalizers ()) {
2058 SGEN_LOG (4, "Finalizer-thread wakeup");
2059 sgen_client_finalize_notify ();
2062 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2063 current_collection_generation = -1;
2065 memset (&counts, 0, sizeof (ScannedObjectCounts));
2066 major_collector.finish_major_collection (&counts);
2068 sgen_workers_assert_gray_queue_is_empty ();
2070 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2071 if (concurrent_collection_in_progress)
2072 concurrent_collection_in_progress = FALSE;
2074 check_scan_starts ();
2076 binary_protocol_flush_buffers (FALSE);
2078 //consistency_check ();
2080 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2084 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2086 TV_DECLARE (time_start);
2087 TV_DECLARE (time_end);
2088 size_t old_next_pin_slot;
2089 SgenGrayQueue gc_thread_gray_queue;
2091 if (disable_major_collections)
2094 if (major_collector.get_and_reset_num_major_objects_marked) {
2095 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2096 g_assert (!num_marked);
2099 /* world must be stopped already */
2100 TV_GETTIME (time_start);
2102 init_gray_queue (&gc_thread_gray_queue, FALSE);
2103 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2104 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2105 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2107 TV_GETTIME (time_end);
2108 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2110 /* FIXME: also report this to the user, preferably in gc-end. */
2111 if (major_collector.get_and_reset_num_major_objects_marked)
2112 major_collector.get_and_reset_num_major_objects_marked ();
2114 return bytes_pinned_from_failed_allocation > 0;
2118 major_start_concurrent_collection (const char *reason)
2120 TV_DECLARE (time_start);
2121 TV_DECLARE (time_end);
2122 long long num_objects_marked;
2123 SgenGrayQueue gc_thread_gray_queue;
2125 if (disable_major_collections)
2128 TV_GETTIME (time_start);
2129 SGEN_TV_GETTIME (time_major_conc_collection_start);
2131 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2132 g_assert (num_objects_marked == 0);
2134 binary_protocol_concurrent_start ();
2136 init_gray_queue (&gc_thread_gray_queue, TRUE);
2137 // FIXME: store reason and pass it when finishing
2138 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2139 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2141 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2143 TV_GETTIME (time_end);
2144 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2146 current_collection_generation = -1;
2150 * Returns whether the major collection has finished.
2153 major_should_finish_concurrent_collection (void)
2155 return sgen_workers_all_done ();
2159 major_update_concurrent_collection (void)
2161 TV_DECLARE (total_start);
2162 TV_DECLARE (total_end);
2164 TV_GETTIME (total_start);
2166 binary_protocol_concurrent_update ();
2168 major_collector.update_cardtable_mod_union ();
2169 sgen_los_update_cardtable_mod_union ();
2171 TV_GETTIME (total_end);
2172 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2176 major_finish_concurrent_collection (gboolean forced)
2178 SgenGrayQueue gc_thread_gray_queue;
2179 TV_DECLARE (total_start);
2180 TV_DECLARE (total_end);
2182 TV_GETTIME (total_start);
2184 binary_protocol_concurrent_finish ();
2187 * We need to stop all workers since we're updating the cardtable below.
2188 * The workers will be resumed with a finishing pause context to avoid
2189 * additional cardtable and object scanning.
2191 sgen_workers_stop_all_workers ();
2193 SGEN_TV_GETTIME (time_major_conc_collection_end);
2194 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2196 major_collector.update_cardtable_mod_union ();
2197 sgen_los_update_cardtable_mod_union ();
2199 if (mod_union_consistency_check)
2200 sgen_check_mod_union_consistency ();
2202 current_collection_generation = GENERATION_OLD;
2203 sgen_cement_reset ();
2204 init_gray_queue (&gc_thread_gray_queue, FALSE);
2205 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2206 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2208 TV_GETTIME (total_end);
2209 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2211 current_collection_generation = -1;
2215 * Ensure an allocation request for @size will succeed by freeing enough memory.
2217 * LOCKING: The GC lock MUST be held.
2220 sgen_ensure_free_space (size_t size, int generation)
2222 int generation_to_collect = -1;
2223 const char *reason = NULL;
2225 if (generation == GENERATION_OLD) {
2226 if (sgen_need_major_collection (size)) {
2227 reason = "LOS overflow";
2228 generation_to_collect = GENERATION_OLD;
2231 if (degraded_mode) {
2232 if (sgen_need_major_collection (size)) {
2233 reason = "Degraded mode overflow";
2234 generation_to_collect = GENERATION_OLD;
2236 } else if (sgen_need_major_collection (size)) {
2237 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2238 generation_to_collect = GENERATION_OLD;
2240 generation_to_collect = GENERATION_NURSERY;
2241 reason = "Nursery full";
2245 if (generation_to_collect == -1) {
2246 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2247 generation_to_collect = GENERATION_OLD;
2248 reason = "Finish concurrent collection";
2252 if (generation_to_collect == -1)
2254 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2258 * LOCKING: Assumes the GC lock is held.
2261 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2263 TV_DECLARE (gc_total_start);
2264 TV_DECLARE (gc_total_end);
2265 int overflow_generation_to_collect = -1;
2266 int oldest_generation_collected = generation_to_collect;
2267 const char *overflow_reason = NULL;
2268 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2270 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2272 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2275 sgen_stop_world (generation_to_collect);
2277 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2280 TV_GETTIME (gc_total_start);
2282 // FIXME: extract overflow reason
2283 // FIXME: minor overflow for concurrent case
2284 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2285 if (concurrent_collection_in_progress)
2286 major_update_concurrent_collection ();
2288 if (collect_nursery (reason, FALSE, NULL) && !concurrent_collection_in_progress) {
2289 overflow_generation_to_collect = GENERATION_OLD;
2290 overflow_reason = "Minor overflow";
2292 } else if (finish_concurrent) {
2293 major_finish_concurrent_collection (wait_to_finish);
2294 oldest_generation_collected = GENERATION_OLD;
2296 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2297 if (major_collector.is_concurrent && !wait_to_finish) {
2298 collect_nursery ("Concurrent start", FALSE, NULL);
2299 major_start_concurrent_collection (reason);
2300 oldest_generation_collected = GENERATION_NURSERY;
2301 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2302 overflow_generation_to_collect = GENERATION_NURSERY;
2303 overflow_reason = "Excessive pinning";
2307 if (overflow_generation_to_collect != -1) {
2308 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2311 * We need to do an overflow collection, either because we ran out of memory
2312 * or the nursery is fully pinned.
2315 if (overflow_generation_to_collect == GENERATION_NURSERY)
2316 collect_nursery (overflow_reason, TRUE, NULL);
2318 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2320 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2323 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2325 /* this also sets the proper pointers for the next allocation */
2326 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2327 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2328 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2329 sgen_dump_pin_queue ();
2333 TV_GETTIME (gc_total_end);
2334 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2337 sgen_restart_world (oldest_generation_collected);
2341 * ######################################################################
2342 * ######## Memory allocation from the OS
2343 * ######################################################################
2344 * This section of code deals with getting memory from the OS and
2345 * allocating memory for GC-internal data structures.
2346 * Internal memory can be handled with a freelist for small objects.
2352 G_GNUC_UNUSED static void
2353 report_internal_mem_usage (void)
2355 printf ("Internal memory usage:\n");
2356 sgen_report_internal_mem_usage ();
2357 printf ("Pinned memory usage:\n");
2358 major_collector.report_pinned_memory_usage ();
2362 * ######################################################################
2363 * ######## Finalization support
2364 * ######################################################################
2368 * If the object has been forwarded it means it's still referenced from a root.
2369 * If it is pinned it's still alive as well.
2370 * A LOS object is only alive if we have pinned it.
2371 * Return TRUE if @obj is ready to be finalized.
2373 static inline gboolean
2374 sgen_is_object_alive (GCObject *object)
2376 if (ptr_in_nursery (object))
2377 return sgen_nursery_is_object_alive (object);
2379 return sgen_major_is_object_alive (object);
2383 * This function returns true if @object is either alive and belongs to the
2384 * current collection - major collections are full heap, so old gen objects
2385 * are never alive during a minor collection.
2388 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2390 if (ptr_in_nursery (object))
2391 return sgen_nursery_is_object_alive (object);
2393 if (current_collection_generation == GENERATION_NURSERY)
2396 return sgen_major_is_object_alive (object);
2401 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2403 return !sgen_is_object_alive (object);
2407 sgen_queue_finalization_entry (GCObject *obj)
2409 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2411 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2413 sgen_client_object_queued_for_finalization (obj);
2417 sgen_object_is_live (GCObject *obj)
2419 return sgen_is_object_alive_and_on_current_collection (obj);
2423 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2424 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2425 * all finalizers have really finished running.
2427 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2428 * This means that just checking whether the queues are empty leaves the possibility that an
2429 * object might have been dequeued but not yet finalized. That's why we need the additional
2430 * flag `pending_unqueued_finalizer`.
2433 static volatile gboolean pending_unqueued_finalizer = FALSE;
2434 volatile gboolean sgen_suspend_finalizers = FALSE;
2437 sgen_set_suspend_finalizers (void)
2439 sgen_suspend_finalizers = TRUE;
2443 sgen_gc_invoke_finalizers (void)
2447 g_assert (!pending_unqueued_finalizer);
2449 /* FIXME: batch to reduce lock contention */
2450 while (sgen_have_pending_finalizers ()) {
2456 * We need to set `pending_unqueued_finalizer` before dequeing the
2457 * finalizable object.
2459 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2460 pending_unqueued_finalizer = TRUE;
2461 mono_memory_write_barrier ();
2462 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2463 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2464 pending_unqueued_finalizer = TRUE;
2465 mono_memory_write_barrier ();
2466 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2472 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2480 /* the object is on the stack so it is pinned */
2481 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2482 sgen_client_run_finalize (obj);
2485 if (pending_unqueued_finalizer) {
2486 mono_memory_write_barrier ();
2487 pending_unqueued_finalizer = FALSE;
2494 sgen_have_pending_finalizers (void)
2496 if (sgen_suspend_finalizers)
2498 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2502 * ######################################################################
2503 * ######## registered roots support
2504 * ######################################################################
2508 * We do not coalesce roots.
2511 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2513 RootRecord new_root;
2516 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2517 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2518 /* we allow changing the size and the descriptor (for thread statics etc) */
2520 size_t old_size = root->end_root - start;
2521 root->end_root = start + size;
2522 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2523 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2524 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2525 root->root_desc = descr;
2527 roots_size -= old_size;
2533 new_root.end_root = start + size;
2534 new_root.root_desc = descr;
2535 new_root.source = source;
2538 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2541 SGEN_LOG (3, "Added root for range: %p-%p, descr: %llx (%d/%d bytes)", start, new_root.end_root, (long long)descr, (int)size, (int)roots_size);
2548 sgen_deregister_root (char* addr)
2554 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2555 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2556 roots_size -= (root.end_root - addr);
2562 * ######################################################################
2563 * ######## Thread handling (stop/start code)
2564 * ######################################################################
2568 sgen_get_current_collection_generation (void)
2570 return current_collection_generation;
2574 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2576 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2578 sgen_client_thread_register (info, stack_bottom_fallback);
2584 sgen_thread_unregister (SgenThreadInfo *p)
2586 sgen_client_thread_unregister (p);
2590 * ######################################################################
2591 * ######## Write barriers
2592 * ######################################################################
2596 * Note: the write barriers first do the needed GC work and then do the actual store:
2597 * this way the value is visible to the conservative GC scan after the write barrier
2598 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2599 * the conservative scan, otherwise by the remembered set scan.
2603 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2605 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2606 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2607 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2608 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2612 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2613 if (binary_protocol_is_heavy_enabled ()) {
2615 for (i = 0; i < count; ++i) {
2616 gpointer dest = (gpointer*)dest_ptr + i;
2617 gpointer obj = *((gpointer*)src_ptr + i);
2619 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2624 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2628 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2632 HEAVY_STAT (++stat_wbarrier_generic_store);
2634 sgen_client_wbarrier_generic_nostore_check (ptr);
2636 obj = *(gpointer*)ptr;
2638 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2641 * We need to record old->old pointer locations for the
2642 * concurrent collector.
2644 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2645 SGEN_LOG (8, "Skipping remset at %p", ptr);
2649 SGEN_LOG (8, "Adding remset at %p", ptr);
2651 remset.wbarrier_generic_nostore (ptr);
2655 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2657 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2658 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2659 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2660 mono_gc_wbarrier_generic_nostore (ptr);
2661 sgen_dummy_use (value);
2664 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2665 * as an atomic operation with release semantics.
2668 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2670 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2672 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2674 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2676 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2677 mono_gc_wbarrier_generic_nostore (ptr);
2679 sgen_dummy_use (value);
2683 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2685 GCObject **dest = (GCObject **)_dest;
2686 GCObject **src = (GCObject **)_src;
2690 mono_gc_wbarrier_generic_store (dest, *src);
2695 size -= SIZEOF_VOID_P;
2701 * ######################################################################
2702 * ######## Other mono public interface functions.
2703 * ######################################################################
2707 sgen_gc_collect (int generation)
2712 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2717 sgen_gc_collection_count (int generation)
2719 if (generation == 0)
2720 return gc_stats.minor_gc_count;
2721 return gc_stats.major_gc_count;
2725 sgen_gc_get_used_size (void)
2729 tot = los_memory_usage;
2730 tot += nursery_section->next_data - nursery_section->data;
2731 tot += major_collector.get_used_size ();
2732 /* FIXME: account for pinned objects */
2738 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2742 va_start (ap, description_format);
2744 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2745 vfprintf (stderr, description_format, ap);
2747 fprintf (stderr, " - %s", fallback);
2748 fprintf (stderr, "\n");
2754 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2757 double val = strtod (opt, &endptr);
2758 if (endptr == opt) {
2759 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2762 else if (val < min || val > max) {
2763 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2775 char *major_collector_opt = NULL;
2776 char *minor_collector_opt = NULL;
2777 char *params_opts = NULL;
2778 char *debug_opts = NULL;
2779 size_t max_heap = 0;
2780 size_t soft_limit = 0;
2782 gboolean debug_print_allowance = FALSE;
2783 double allowance_ratio = 0, save_target = 0;
2784 gboolean cement_enabled = TRUE;
2787 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2790 /* already inited */
2793 /* being inited by another thread */
2794 mono_thread_info_usleep (1000);
2797 /* we will init it */
2800 g_assert_not_reached ();
2802 } while (result != 0);
2804 SGEN_TV_GETTIME (sgen_init_timestamp);
2806 #ifdef SGEN_WITHOUT_MONO
2807 mono_thread_smr_init ();
2810 mono_coop_mutex_init (&gc_mutex);
2812 gc_debug_file = stderr;
2814 mono_coop_mutex_init (&sgen_interruption_mutex);
2816 if ((env = g_getenv (MONO_GC_PARAMS_NAME)) || gc_params_options) {
2817 params_opts = g_strdup_printf ("%s,%s", gc_params_options ? gc_params_options : "", env ? env : "");
2821 opts = g_strsplit (params_opts, ",", -1);
2822 for (ptr = opts; *ptr; ++ptr) {
2824 if (g_str_has_prefix (opt, "major=")) {
2825 opt = strchr (opt, '=') + 1;
2826 major_collector_opt = g_strdup (opt);
2827 } else if (g_str_has_prefix (opt, "minor=")) {
2828 opt = strchr (opt, '=') + 1;
2829 minor_collector_opt = g_strdup (opt);
2837 sgen_init_internal_allocator ();
2838 sgen_init_nursery_allocator ();
2839 sgen_init_fin_weak_hash ();
2840 sgen_init_hash_table ();
2841 sgen_init_descriptors ();
2842 sgen_init_gray_queues ();
2843 sgen_init_allocator ();
2844 sgen_init_gchandles ();
2846 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2847 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2849 sgen_client_init ();
2851 if (!minor_collector_opt) {
2852 sgen_simple_nursery_init (&sgen_minor_collector);
2854 if (!strcmp (minor_collector_opt, "simple")) {
2856 sgen_simple_nursery_init (&sgen_minor_collector);
2857 } else if (!strcmp (minor_collector_opt, "split")) {
2858 sgen_split_nursery_init (&sgen_minor_collector);
2860 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2861 goto use_simple_nursery;
2865 if (!major_collector_opt) {
2867 DEFAULT_MAJOR_INIT (&major_collector);
2868 } else if (!strcmp (major_collector_opt, "marksweep")) {
2869 sgen_marksweep_init (&major_collector);
2870 } else if (!strcmp (major_collector_opt, "marksweep-conc")) {
2871 sgen_marksweep_conc_init (&major_collector);
2872 } else if (!strcmp (major_collector_opt, "marksweep-conc-par")) {
2873 sgen_marksweep_conc_par_init (&major_collector);
2875 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `" DEFAULT_MAJOR_NAME "` instead.", "Unknown major collector `%s'.", major_collector_opt);
2876 goto use_default_major;
2879 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2882 gboolean usage_printed = FALSE;
2884 for (ptr = opts; *ptr; ++ptr) {
2886 if (!strcmp (opt, ""))
2888 if (g_str_has_prefix (opt, "major="))
2890 if (g_str_has_prefix (opt, "minor="))
2892 if (g_str_has_prefix (opt, "max-heap-size=")) {
2893 size_t page_size = mono_pagesize ();
2894 size_t max_heap_candidate = 0;
2895 opt = strchr (opt, '=') + 1;
2896 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2897 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2898 if (max_heap != max_heap_candidate)
2899 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2901 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2905 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2906 opt = strchr (opt, '=') + 1;
2907 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2908 if (soft_limit <= 0) {
2909 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2913 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2919 if (g_str_has_prefix (opt, "nursery-size=")) {
2921 opt = strchr (opt, '=') + 1;
2922 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2923 if ((val & (val - 1))) {
2924 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2928 if (val < SGEN_MAX_NURSERY_WASTE) {
2929 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2930 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2934 sgen_nursery_size = val;
2935 sgen_nursery_bits = 0;
2936 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2939 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2945 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2947 opt = strchr (opt, '=') + 1;
2948 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2949 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2954 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2956 opt = strchr (opt, '=') + 1;
2957 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2958 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2959 allowance_ratio = val;
2964 if (!strcmp (opt, "cementing")) {
2965 cement_enabled = TRUE;
2968 if (!strcmp (opt, "no-cementing")) {
2969 cement_enabled = FALSE;
2973 if (!strcmp (opt, "precleaning")) {
2974 precleaning_enabled = TRUE;
2977 if (!strcmp (opt, "no-precleaning")) {
2978 precleaning_enabled = FALSE;
2982 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2985 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2988 if (sgen_client_handle_gc_param (opt))
2991 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2996 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2997 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2998 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2999 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3000 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
3001 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
3002 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
3003 fprintf (stderr, " [no-]cementing\n");
3004 if (major_collector.print_gc_param_usage)
3005 major_collector.print_gc_param_usage ();
3006 if (sgen_minor_collector.print_gc_param_usage)
3007 sgen_minor_collector.print_gc_param_usage ();
3008 sgen_client_print_gc_params_usage ();
3009 fprintf (stderr, " Experimental options:\n");
3010 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3011 fprintf (stderr, " default-allowance-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO);
3012 fprintf (stderr, "\n");
3014 usage_printed = TRUE;
3019 if (major_collector_opt)
3020 g_free (major_collector_opt);
3022 if (minor_collector_opt)
3023 g_free (minor_collector_opt);
3026 g_free (params_opts);
3030 sgen_pinning_init ();
3031 sgen_cement_init (cement_enabled);
3033 if ((env = g_getenv (MONO_GC_DEBUG_NAME)) || gc_debug_options) {
3034 debug_opts = g_strdup_printf ("%s,%s", gc_debug_options ? gc_debug_options : "", env ? env : "");
3038 gboolean usage_printed = FALSE;
3040 opts = g_strsplit (debug_opts, ",", -1);
3041 for (ptr = opts; ptr && *ptr; ptr ++) {
3043 if (!strcmp (opt, ""))
3045 if (opt [0] >= '0' && opt [0] <= '9') {
3046 gc_debug_level = atoi (opt);
3051 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3052 gc_debug_file = fopen (rf, "wb");
3054 gc_debug_file = stderr;
3057 } else if (!strcmp (opt, "print-allowance")) {
3058 debug_print_allowance = TRUE;
3059 } else if (!strcmp (opt, "print-pinning")) {
3060 sgen_pin_stats_enable ();
3061 } else if (!strcmp (opt, "verify-before-allocs")) {
3062 verify_before_allocs = 1;
3063 has_per_allocation_action = TRUE;
3064 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3065 char *arg = strchr (opt, '=') + 1;
3066 verify_before_allocs = atoi (arg);
3067 has_per_allocation_action = TRUE;
3068 } else if (!strcmp (opt, "collect-before-allocs")) {
3069 collect_before_allocs = 1;
3070 has_per_allocation_action = TRUE;
3071 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3072 char *arg = strchr (opt, '=') + 1;
3073 has_per_allocation_action = TRUE;
3074 collect_before_allocs = atoi (arg);
3075 } else if (!strcmp (opt, "verify-before-collections")) {
3076 whole_heap_check_before_collection = TRUE;
3077 } else if (!strcmp (opt, "check-remset-consistency")) {
3078 remset_consistency_checks = TRUE;
3079 nursery_clear_policy = CLEAR_AT_GC;
3080 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3081 if (!major_collector.is_concurrent) {
3082 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3085 mod_union_consistency_check = TRUE;
3086 } else if (!strcmp (opt, "check-mark-bits")) {
3087 check_mark_bits_after_major_collection = TRUE;
3088 } else if (!strcmp (opt, "check-nursery-pinned")) {
3089 check_nursery_objects_pinned = TRUE;
3090 } else if (!strcmp (opt, "clear-at-gc")) {
3091 nursery_clear_policy = CLEAR_AT_GC;
3092 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3093 nursery_clear_policy = CLEAR_AT_GC;
3094 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3095 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3096 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3097 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3098 } else if (!strcmp (opt, "check-scan-starts")) {
3099 do_scan_starts_check = TRUE;
3100 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3101 do_verify_nursery = TRUE;
3102 } else if (!strcmp (opt, "check-concurrent")) {
3103 if (!major_collector.is_concurrent) {
3104 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3107 nursery_clear_policy = CLEAR_AT_GC;
3108 do_concurrent_checks = TRUE;
3109 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3110 do_dump_nursery_content = TRUE;
3111 } else if (!strcmp (opt, "disable-minor")) {
3112 disable_minor_collections = TRUE;
3113 } else if (!strcmp (opt, "disable-major")) {
3114 disable_major_collections = TRUE;
3115 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3116 char *filename = strchr (opt, '=') + 1;
3117 nursery_clear_policy = CLEAR_AT_GC;
3118 sgen_debug_enable_heap_dump (filename);
3119 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3120 char *filename = strchr (opt, '=') + 1;
3121 char *colon = strrchr (filename, ':');
3124 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3125 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3130 binary_protocol_init (filename, (long long)limit);
3131 } else if (!strcmp (opt, "nursery-canaries")) {
3132 do_verify_nursery = TRUE;
3133 enable_nursery_canaries = TRUE;
3134 } else if (!sgen_client_handle_gc_debug (opt)) {
3135 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3140 fprintf (stderr, "\n%s must be of the format [<l>[:<filename>]|<option>]+ where <l> is a debug level 0-9.\n", MONO_GC_DEBUG_NAME);
3141 fprintf (stderr, "Valid <option>s are:\n");
3142 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3143 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3144 fprintf (stderr, " check-remset-consistency\n");
3145 fprintf (stderr, " check-mark-bits\n");
3146 fprintf (stderr, " check-nursery-pinned\n");
3147 fprintf (stderr, " verify-before-collections\n");
3148 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3149 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3150 fprintf (stderr, " disable-minor\n");
3151 fprintf (stderr, " disable-major\n");
3152 fprintf (stderr, " check-concurrent\n");
3153 fprintf (stderr, " clear-[nursery-]at-gc\n");
3154 fprintf (stderr, " clear-at-tlab-creation\n");
3155 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3156 fprintf (stderr, " check-scan-starts\n");
3157 fprintf (stderr, " print-allowance\n");
3158 fprintf (stderr, " print-pinning\n");
3159 fprintf (stderr, " heap-dump=<filename>\n");
3160 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3161 fprintf (stderr, " nursery-canaries\n");
3162 sgen_client_print_gc_debug_usage ();
3163 fprintf (stderr, "\n");
3165 usage_printed = TRUE;
3172 g_free (debug_opts);
3174 if (check_mark_bits_after_major_collection)
3175 nursery_clear_policy = CLEAR_AT_GC;
3177 if (major_collector.post_param_init)
3178 major_collector.post_param_init (&major_collector);
3180 if (major_collector.needs_thread_pool) {
3181 int num_workers = 1;
3182 if (major_collector.is_parallel) {
3183 /* FIXME Detect the number of physical cores, instead of logical */
3184 num_workers = mono_cpu_count () / 2;
3185 if (num_workers < 1)
3188 sgen_workers_init (num_workers);
3191 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3193 memset (&remset, 0, sizeof (remset));
3195 sgen_card_table_init (&remset);
3197 sgen_register_root (NULL, 0, sgen_make_user_root_descriptor (sgen_mark_normal_gc_handles), ROOT_TYPE_NORMAL, MONO_ROOT_SOURCE_GC_HANDLE, "normal gc handles");
3201 sgen_init_bridge ();
3205 sgen_gc_initialized ()
3207 return gc_initialized > 0;
3211 sgen_get_nursery_clear_policy (void)
3213 return nursery_clear_policy;
3219 mono_coop_mutex_lock (&gc_mutex);
3223 sgen_gc_unlock (void)
3225 mono_coop_mutex_unlock (&gc_mutex);
3229 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3231 major_collector.iterate_live_block_ranges (callback);
3235 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3237 major_collector.iterate_block_ranges (callback);
3241 sgen_get_major_collector (void)
3243 return &major_collector;
3247 sgen_get_remset (void)
3253 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3255 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3256 sgen_los_count_cards (los_total, los_marked);
3259 static gboolean world_is_stopped = FALSE;
3261 /* LOCKING: assumes the GC lock is held */
3263 sgen_stop_world (int generation)
3265 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3267 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3269 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3271 sgen_client_stop_world (generation);
3273 world_is_stopped = TRUE;
3275 if (binary_protocol_is_heavy_enabled ())
3276 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3277 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3280 /* LOCKING: assumes the GC lock is held */
3282 sgen_restart_world (int generation)
3284 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3287 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3289 if (binary_protocol_is_heavy_enabled ())
3290 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3291 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3293 world_is_stopped = FALSE;
3295 sgen_client_restart_world (generation, &stw_time);
3297 binary_protocol_world_restarted (generation, sgen_timestamp ());
3299 if (sgen_client_bridge_need_processing ())
3300 sgen_client_bridge_processing_finish (generation);
3302 sgen_memgov_collection_end (generation, stw_time);
3306 sgen_is_world_stopped (void)
3308 return world_is_stopped;
3312 sgen_check_whole_heap_stw (void)
3314 sgen_stop_world (0);
3315 sgen_clear_nursery_fragments ();
3316 sgen_check_whole_heap (TRUE);
3317 sgen_restart_world (0);
3321 sgen_timestamp (void)
3323 SGEN_TV_DECLARE (timestamp);
3324 SGEN_TV_GETTIME (timestamp);
3325 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3328 #endif /* HAVE_SGEN_GC */