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;
296 mono_gc_flush_info (void)
298 fflush (gc_debug_file);
302 #define TV_DECLARE SGEN_TV_DECLARE
303 #define TV_GETTIME SGEN_TV_GETTIME
304 #define TV_ELAPSED SGEN_TV_ELAPSED
306 static SGEN_TV_DECLARE (sgen_init_timestamp);
308 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
310 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
311 #define object_is_pinned SGEN_OBJECT_IS_PINNED
312 #define pin_object SGEN_PIN_OBJECT
314 #define ptr_in_nursery sgen_ptr_in_nursery
316 #define LOAD_VTABLE SGEN_LOAD_VTABLE
319 nursery_canaries_enabled (void)
321 return enable_nursery_canaries;
324 #define safe_object_get_size sgen_safe_object_get_size
326 #if defined(HAVE_CONC_GC_AS_DEFAULT)
327 /* Use concurrent major on deskstop platforms */
328 #define DEFAULT_MAJOR_INIT sgen_marksweep_conc_init
329 #define DEFAULT_MAJOR_NAME "marksweep-conc"
331 #define DEFAULT_MAJOR_INIT sgen_marksweep_init
332 #define DEFAULT_MAJOR_NAME "marksweep"
336 * ######################################################################
337 * ######## Global data.
338 * ######################################################################
340 MonoCoopMutex gc_mutex;
342 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
344 size_t degraded_mode = 0;
346 static mword bytes_pinned_from_failed_allocation = 0;
348 GCMemSection *nursery_section = NULL;
349 static volatile mword lowest_heap_address = ~(mword)0;
350 static volatile mword highest_heap_address = 0;
352 MonoCoopMutex sgen_interruption_mutex;
354 int current_collection_generation = -1;
355 static volatile gboolean concurrent_collection_in_progress = FALSE;
357 /* objects that are ready to be finalized */
358 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
359 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
361 /* registered roots: the key to the hash is the root start address */
363 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
365 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
366 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
367 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
368 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
370 static mword roots_size = 0; /* amount of memory in the root set */
372 /* The size of a TLAB */
373 /* The bigger the value, the less often we have to go to the slow path to allocate a new
374 * one, but the more space is wasted by threads not allocating much memory.
376 * FIXME: Make this self-tuning for each thread.
378 guint32 tlab_size = (1024 * 4);
380 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
382 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
384 #define ALIGN_UP SGEN_ALIGN_UP
386 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
387 MonoNativeThreadId main_gc_thread = NULL;
390 /*Object was pinned during the current collection*/
391 static mword objects_pinned;
394 * ######################################################################
395 * ######## Macros and function declarations.
396 * ######################################################################
399 /* forward declarations */
400 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
402 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
403 static void finish_gray_stack (int generation, ScanCopyContext ctx);
406 SgenMajorCollector major_collector;
407 SgenMinorCollector sgen_minor_collector;
409 static SgenRememberedSet remset;
412 * The gray queue a worker job must use. If we're not parallel or
413 * concurrent, we use the main gray queue.
415 static SgenGrayQueue*
416 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
419 return &worker_data->private_gray_queue;
420 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
421 return default_gray_queue;
425 gray_queue_enable_redirect (SgenGrayQueue *queue)
427 SGEN_ASSERT (0, concurrent_collection_in_progress, "Where are we redirecting the gray queue to, without a concurrent collection?");
429 sgen_gray_queue_set_alloc_prepare (queue, sgen_workers_take_from_queue_and_awake);
430 sgen_workers_take_from_queue_and_awake (queue);
434 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
436 while (start < end) {
440 if (!*(void**)start) {
441 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
446 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
452 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
453 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
454 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
455 callback ((GCObject*)obj, size, data);
456 CANARIFY_SIZE (size);
458 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
466 * sgen_add_to_global_remset:
468 * The global remset contains locations which point into newspace after
469 * a minor collection. This can happen if the objects they point to are pinned.
471 * LOCKING: If called from a parallel collector, the global remset
472 * lock must be held. For serial collectors that is not necessary.
475 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
477 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
479 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
481 if (!major_collector.is_concurrent) {
482 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
484 if (current_collection_generation == -1)
485 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
488 if (!object_is_pinned (obj))
489 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");
490 else if (sgen_cement_lookup_or_register (obj))
493 remset.record_pointer (ptr);
495 sgen_pin_stats_register_global_remset (obj);
497 SGEN_LOG (8, "Adding global remset for %p", ptr);
498 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
502 * sgen_drain_gray_stack:
504 * Scan objects in the gray stack until the stack is empty. This should be called
505 * frequently after each object is copied, to achieve better locality and cache
510 sgen_drain_gray_stack (ScanCopyContext ctx)
512 ScanObjectFunc scan_func = ctx.ops->scan_object;
513 SgenGrayQueue *queue = ctx.queue;
515 if (ctx.ops->drain_gray_stack)
516 return ctx.ops->drain_gray_stack (queue);
521 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
524 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
525 scan_func (obj, desc, queue);
531 * Addresses in the pin queue are already sorted. This function finds
532 * the object header for each address and pins the object. The
533 * addresses must be inside the nursery section. The (start of the)
534 * address array is overwritten with the addresses of the actually
535 * pinned objects. Return the number of pinned objects.
538 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
540 GCMemSection *section = nursery_section;
541 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
542 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
543 void *start_nursery = section->data;
544 void *end_nursery = section->next_data;
549 void *pinning_front = start_nursery;
551 void **definitely_pinned = start;
552 ScanObjectFunc scan_func = ctx.ops->scan_object;
553 SgenGrayQueue *queue = ctx.queue;
555 sgen_nursery_allocator_prepare_for_pinning ();
557 while (start < end) {
558 GCObject *obj_to_pin = NULL;
559 size_t obj_to_pin_size = 0;
564 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
565 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
572 SGEN_LOG (5, "Considering pinning addr %p", addr);
573 /* We've already processed everything up to pinning_front. */
574 if (addr < pinning_front) {
580 * Find the closest scan start <= addr. We might search backward in the
581 * scan_starts array because entries might be NULL. In the worst case we
582 * start at start_nursery.
584 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
585 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
586 search_start = (void*)section->scan_starts [idx];
587 if (!search_start || search_start > addr) {
590 search_start = section->scan_starts [idx];
591 if (search_start && search_start <= addr)
594 if (!search_start || search_start > addr)
595 search_start = start_nursery;
599 * If the pinning front is closer than the scan start we found, start
600 * searching at the front.
602 if (search_start < pinning_front)
603 search_start = pinning_front;
606 * Now addr should be in an object a short distance from search_start.
608 * search_start must point to zeroed mem or point to an object.
611 size_t obj_size, canarified_obj_size;
614 if (!*(void**)search_start) {
615 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
616 /* The loop condition makes sure we don't overrun addr. */
620 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
623 * Filler arrays are marked by an invalid sync word. We don't
624 * consider them for pinning. They are not delimited by canaries,
627 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
628 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
629 CANARIFY_SIZE (canarified_obj_size);
631 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
632 /* This is the object we're looking for. */
633 obj_to_pin = (GCObject*)search_start;
634 obj_to_pin_size = canarified_obj_size;
639 /* Skip to the next object */
640 search_start = (void*)((char*)search_start + canarified_obj_size);
641 } while (search_start <= addr);
643 /* We've searched past the address we were looking for. */
645 pinning_front = search_start;
646 goto next_pin_queue_entry;
650 * We've found an object to pin. It might still be a dummy array, but we
651 * can advance the pinning front in any case.
653 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
656 * If this is a dummy array marking the beginning of a nursery
657 * fragment, we don't pin it.
659 if (sgen_client_object_is_array_fill (obj_to_pin))
660 goto next_pin_queue_entry;
663 * Finally - pin the object!
665 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
666 if (do_scan_objects) {
667 scan_func (obj_to_pin, desc, queue);
669 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
670 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
671 binary_protocol_pin (obj_to_pin,
672 (gpointer)LOAD_VTABLE (obj_to_pin),
673 safe_object_get_size (obj_to_pin));
675 pin_object (obj_to_pin);
676 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
677 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
678 definitely_pinned [count] = obj_to_pin;
681 if (concurrent_collection_in_progress)
682 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
684 next_pin_queue_entry:
688 sgen_client_nursery_objects_pinned (definitely_pinned, count);
689 stat_pinned_objects += count;
694 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
698 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
701 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
702 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
706 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
707 * when we can't promote an object because we're out of memory.
710 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
712 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
715 * All pinned objects are assumed to have been staged, so we need to stage as well.
716 * Also, the count of staged objects shows that "late pinning" happened.
718 sgen_pin_stage_ptr (object);
720 SGEN_PIN_OBJECT (object);
721 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
724 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
726 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
729 /* Sort the addresses in array in increasing order.
730 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
733 sgen_sort_addresses (void **array, size_t size)
738 for (i = 1; i < size; ++i) {
741 size_t parent = (child - 1) / 2;
743 if (array [parent] >= array [child])
746 tmp = array [parent];
747 array [parent] = array [child];
754 for (i = size - 1; i > 0; --i) {
757 array [i] = array [0];
763 while (root * 2 + 1 <= end) {
764 size_t child = root * 2 + 1;
766 if (child < end && array [child] < array [child + 1])
768 if (array [root] >= array [child])
772 array [root] = array [child];
781 * Scan the memory between start and end and queue values which could be pointers
782 * to the area between start_nursery and end_nursery for later consideration.
783 * Typically used for thread stacks.
786 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
790 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
792 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
793 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
796 while (start < end) {
798 * *start can point to the middle of an object
799 * note: should we handle pointing at the end of an object?
800 * pinning in C# code disallows pointing at the end of an object
801 * but there is some small chance that an optimizing C compiler
802 * may keep the only reference to an object by pointing
803 * at the end of it. We ignore this small chance for now.
804 * Pointers to the end of an object are indistinguishable
805 * from pointers to the start of the next object in memory
806 * so if we allow that we'd need to pin two objects...
807 * We queue the pointer in an array, the
808 * array will then be sorted and uniqued. This way
809 * we can coalesce several pinning pointers and it should
810 * be faster since we'd do a memory scan with increasing
811 * addresses. Note: we can align the address to the allocation
812 * alignment, so the unique process is more effective.
814 mword addr = (mword)*start;
815 addr &= ~(ALLOC_ALIGN - 1);
816 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
817 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
818 sgen_pin_stage_ptr ((void*)addr);
819 binary_protocol_pin_stage (start, (void*)addr);
820 sgen_pin_stats_register_address ((char*)addr, pin_type);
826 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
830 * The first thing we do in a collection is to identify pinned objects.
831 * This function considers all the areas of memory that need to be
832 * conservatively scanned.
835 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
839 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);
840 /* objects pinned from the API are inside these roots */
841 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
842 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
843 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
844 } SGEN_HASH_TABLE_FOREACH_END;
845 /* now deal with the thread stacks
846 * in the future we should be able to conservatively scan only:
847 * *) the cpu registers
848 * *) the unmanaged stack frames
849 * *) the _last_ managed stack frame
850 * *) pointers slots in managed frames
852 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
856 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
858 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
859 ctx->ops->copy_or_mark_object (obj, ctx->queue);
863 * The memory area from start_root to end_root contains pointers to objects.
864 * Their position is precisely described by @desc (this means that the pointer
865 * can be either NULL or the pointer to the start of an object).
866 * This functions copies them to to_space updates them.
868 * This function is not thread-safe!
871 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
873 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
874 SgenGrayQueue *queue = ctx.queue;
876 switch (desc & ROOT_DESC_TYPE_MASK) {
877 case ROOT_DESC_BITMAP:
878 desc >>= ROOT_DESC_TYPE_SHIFT;
880 if ((desc & 1) && *start_root) {
881 copy_func ((GCObject**)start_root, queue);
882 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
888 case ROOT_DESC_COMPLEX: {
889 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
890 gsize bwords = (*bitmap_data) - 1;
891 void **start_run = start_root;
893 while (bwords-- > 0) {
894 gsize bmap = *bitmap_data++;
895 void **objptr = start_run;
897 if ((bmap & 1) && *objptr) {
898 copy_func ((GCObject**)objptr, queue);
899 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
904 start_run += GC_BITS_PER_WORD;
908 case ROOT_DESC_USER: {
909 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
910 marker (start_root, single_arg_user_copy_or_mark, &ctx);
913 case ROOT_DESC_RUN_LEN:
914 g_assert_not_reached ();
916 g_assert_not_reached ();
921 reset_heap_boundaries (void)
923 lowest_heap_address = ~(mword)0;
924 highest_heap_address = 0;
928 sgen_update_heap_boundaries (mword low, mword high)
933 old = lowest_heap_address;
936 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
939 old = highest_heap_address;
942 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
946 * Allocate and setup the data structures needed to be able to allocate objects
947 * in the nursery. The nursery is stored in nursery_section.
952 GCMemSection *section;
959 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
960 /* later we will alloc a larger area for the nursery but only activate
961 * what we need. The rest will be used as expansion if we have too many pinned
962 * objects in the existing nursery.
964 /* FIXME: handle OOM */
965 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
967 alloc_size = sgen_nursery_size;
969 /* If there isn't enough space even for the nursery we should simply abort. */
970 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
972 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
973 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
974 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 ());
975 section->data = section->next_data = data;
976 section->size = alloc_size;
977 section->end_data = data + sgen_nursery_size;
978 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
979 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
980 section->num_scan_start = scan_starts;
982 nursery_section = section;
984 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
988 mono_gc_get_logfile (void)
990 return gc_debug_file;
994 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
996 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
997 SgenGrayQueue *queue = ctx.queue;
1000 for (i = 0; i < fin_queue->next_slot; ++i) {
1001 GCObject *obj = (GCObject *)fin_queue->data [i];
1004 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1005 copy_func ((GCObject**)&fin_queue->data [i], queue);
1010 generation_name (int generation)
1012 switch (generation) {
1013 case GENERATION_NURSERY: return "nursery";
1014 case GENERATION_OLD: return "old";
1015 default: g_assert_not_reached ();
1020 sgen_generation_name (int generation)
1022 return generation_name (generation);
1026 finish_gray_stack (int generation, ScanCopyContext ctx)
1030 int done_with_ephemerons, ephemeron_rounds = 0;
1031 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1032 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1033 SgenGrayQueue *queue = ctx.queue;
1035 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1037 * We copied all the reachable objects. Now it's the time to copy
1038 * the objects that were not referenced by the roots, but by the copied objects.
1039 * we built a stack of objects pointed to by gray_start: they are
1040 * additional roots and we may add more items as we go.
1041 * We loop until gray_start == gray_objects which means no more objects have
1042 * been added. Note this is iterative: no recursion is involved.
1043 * We need to walk the LO list as well in search of marked big objects
1044 * (use a flag since this is needed only on major collections). We need to loop
1045 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1046 * To achieve better cache locality and cache usage, we drain the gray stack
1047 * frequently, after each object is copied, and just finish the work here.
1049 sgen_drain_gray_stack (ctx);
1051 SGEN_LOG (2, "%s generation done", generation_name (generation));
1054 Reset bridge data, we might have lingering data from a previous collection if this is a major
1055 collection trigged by minor overflow.
1057 We must reset the gathered bridges since their original block might be evacuated due to major
1058 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1060 if (sgen_client_bridge_need_processing ())
1061 sgen_client_bridge_reset_data ();
1064 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1065 * to ensure they see the full set of live objects.
1067 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1070 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1071 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1072 * objects that are in fact reachable.
1074 done_with_ephemerons = 0;
1076 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1077 sgen_drain_gray_stack (ctx);
1079 } while (!done_with_ephemerons);
1081 if (sgen_client_bridge_need_processing ()) {
1082 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1083 sgen_drain_gray_stack (ctx);
1084 sgen_collect_bridge_objects (generation, ctx);
1085 if (generation == GENERATION_OLD)
1086 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1089 Do the first bridge step here, as the collector liveness state will become useless after that.
1091 An important optimization is to only proccess the possibly dead part of the object graph and skip
1092 over all live objects as we transitively know everything they point must be alive too.
1094 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1096 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1097 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1100 sgen_client_bridge_processing_stw_step ();
1104 Make sure we drain the gray stack before processing disappearing links and finalizers.
1105 If we don't make sure it is empty we might wrongly see a live object as dead.
1107 sgen_drain_gray_stack (ctx);
1110 We must clear weak links that don't track resurrection before processing object ready for
1111 finalization so they can be cleared before that.
1113 sgen_null_link_in_range (generation, ctx, FALSE);
1114 if (generation == GENERATION_OLD)
1115 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1118 /* walk the finalization queue and move also the objects that need to be
1119 * finalized: use the finalized objects as new roots so the objects they depend
1120 * on are also not reclaimed. As with the roots above, only objects in the nursery
1121 * are marked/copied.
1123 sgen_finalize_in_range (generation, ctx);
1124 if (generation == GENERATION_OLD)
1125 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1126 /* drain the new stack that might have been created */
1127 SGEN_LOG (6, "Precise scan of gray area post fin");
1128 sgen_drain_gray_stack (ctx);
1131 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1133 done_with_ephemerons = 0;
1135 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1136 sgen_drain_gray_stack (ctx);
1138 } while (!done_with_ephemerons);
1140 sgen_client_clear_unreachable_ephemerons (ctx);
1143 * We clear togglerefs only after all possible chances of revival are done.
1144 * This is semantically more inline with what users expect and it allows for
1145 * user finalizers to correctly interact with TR objects.
1147 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1150 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);
1153 * handle disappearing links
1154 * Note we do this after checking the finalization queue because if an object
1155 * survives (at least long enough to be finalized) we don't clear the link.
1156 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1157 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1160 g_assert (sgen_gray_object_queue_is_empty (queue));
1162 sgen_null_link_in_range (generation, ctx, TRUE);
1163 if (generation == GENERATION_OLD)
1164 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1165 if (sgen_gray_object_queue_is_empty (queue))
1167 sgen_drain_gray_stack (ctx);
1170 g_assert (sgen_gray_object_queue_is_empty (queue));
1172 sgen_gray_object_queue_trim_free_list (queue);
1173 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1177 sgen_check_section_scan_starts (GCMemSection *section)
1180 for (i = 0; i < section->num_scan_start; ++i) {
1181 if (section->scan_starts [i]) {
1182 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1183 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1189 check_scan_starts (void)
1191 if (!do_scan_starts_check)
1193 sgen_check_section_scan_starts (nursery_section);
1194 major_collector.check_scan_starts ();
1198 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1202 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1203 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1204 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1205 } SGEN_HASH_TABLE_FOREACH_END;
1211 static gboolean inited = FALSE;
1216 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1218 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1219 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1220 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1221 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1222 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1223 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1225 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1226 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1227 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1228 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1229 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1230 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1231 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1232 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1233 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1234 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1236 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1238 #ifdef HEAVY_STATISTICS
1239 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1240 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1241 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1242 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1243 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1245 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1246 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1248 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1249 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1250 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1251 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1253 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1254 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1256 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1258 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1259 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1260 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1261 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1263 sgen_nursery_allocator_init_heavy_stats ();
1271 reset_pinned_from_failed_allocation (void)
1273 bytes_pinned_from_failed_allocation = 0;
1277 sgen_set_pinned_from_failed_allocation (mword objsize)
1279 bytes_pinned_from_failed_allocation += objsize;
1283 sgen_collection_is_concurrent (void)
1285 switch (current_collection_generation) {
1286 case GENERATION_NURSERY:
1288 case GENERATION_OLD:
1289 return concurrent_collection_in_progress;
1291 g_error ("Invalid current generation %d", current_collection_generation);
1297 sgen_concurrent_collection_in_progress (void)
1299 return concurrent_collection_in_progress;
1303 SgenThreadPoolJob job;
1304 SgenObjectOperations *ops;
1305 SgenGrayQueue *gc_thread_gray_queue;
1308 static ScanCopyContext
1309 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1311 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1313 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1317 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1319 remset.scan_remsets (scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job));
1327 } ScanFromRegisteredRootsJob;
1330 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1332 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1333 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1335 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1342 } ScanThreadDataJob;
1345 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1347 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1348 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1350 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1355 SgenPointerQueue *queue;
1356 } ScanFinalizerEntriesJob;
1359 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1361 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1362 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1364 scan_finalizer_entries (job_data->queue, ctx);
1368 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1370 ScanJob *job_data = (ScanJob*)job;
1371 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1373 g_assert (concurrent_collection_in_progress);
1374 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1378 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1380 ScanJob *job_data = (ScanJob*)job;
1381 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1383 g_assert (concurrent_collection_in_progress);
1384 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1388 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1390 ScanJob *job_data = (ScanJob*)job;
1391 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1393 g_assert (concurrent_collection_in_progress);
1395 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1396 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1398 sgen_scan_pin_queue_objects (ctx);
1402 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue, gboolean use_workers)
1405 sgen_workers_init_distribute_gray_queue ();
1406 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1410 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1412 ScanFromRegisteredRootsJob *scrrj;
1413 ScanThreadDataJob *stdj;
1414 ScanFinalizerEntriesJob *sfej;
1416 /* registered roots, this includes static fields */
1418 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1419 scrrj->scan_job.ops = ops;
1420 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1421 scrrj->heap_start = heap_start;
1422 scrrj->heap_end = heap_end;
1423 scrrj->root_type = ROOT_TYPE_NORMAL;
1424 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1426 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1427 scrrj->scan_job.ops = ops;
1428 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1429 scrrj->heap_start = heap_start;
1430 scrrj->heap_end = heap_end;
1431 scrrj->root_type = ROOT_TYPE_WBARRIER;
1432 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1436 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1437 stdj->scan_job.ops = ops;
1438 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1439 stdj->heap_start = heap_start;
1440 stdj->heap_end = heap_end;
1441 sgen_workers_enqueue_job (&stdj->scan_job.job, enqueue);
1443 /* Scan the list of objects ready for finalization. */
1445 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1446 sfej->scan_job.ops = ops;
1447 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1448 sfej->queue = &fin_ready_queue;
1449 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1451 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1452 sfej->scan_job.ops = ops;
1453 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1454 sfej->queue = &critical_fin_queue;
1455 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1459 * Perform a nursery collection.
1461 * Return whether any objects were late-pinned due to being out of memory.
1464 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1466 gboolean needs_major;
1467 size_t max_garbage_amount;
1469 mword fragment_total;
1471 SgenGrayQueue gc_thread_gray_queue;
1472 SgenObjectOperations *object_ops;
1473 ScanCopyContext ctx;
1476 SGEN_TV_DECLARE (last_minor_collection_start_tv);
1477 SGEN_TV_DECLARE (last_minor_collection_end_tv);
1479 if (disable_minor_collections)
1482 TV_GETTIME (last_minor_collection_start_tv);
1483 atv = last_minor_collection_start_tv;
1485 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1487 if (sgen_concurrent_collection_in_progress ())
1488 object_ops = &sgen_minor_collector.serial_ops_with_concurrent_major;
1490 object_ops = &sgen_minor_collector.serial_ops;
1492 if (do_verify_nursery || do_dump_nursery_content)
1493 sgen_debug_verify_nursery (do_dump_nursery_content);
1495 current_collection_generation = GENERATION_NURSERY;
1497 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1499 reset_pinned_from_failed_allocation ();
1501 check_scan_starts ();
1503 sgen_nursery_alloc_prepare_for_minor ();
1507 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1508 /* FIXME: optimize later to use the higher address where an object can be present */
1509 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1511 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 ()));
1512 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1513 g_assert (nursery_section->size >= max_garbage_amount);
1515 /* world must be stopped already */
1517 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1519 sgen_client_pre_collection_checks ();
1521 nursery_section->next_data = nursery_next;
1523 major_collector.start_nursery_collection ();
1525 sgen_memgov_minor_collection_start ();
1527 init_gray_queue (&gc_thread_gray_queue, FALSE);
1528 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gc_thread_gray_queue);
1530 gc_stats.minor_gc_count ++;
1532 if (whole_heap_check_before_collection) {
1533 sgen_clear_nursery_fragments ();
1534 sgen_check_whole_heap (finish_up_concurrent_mark);
1537 sgen_process_fin_stage_entries ();
1539 /* pin from pinned handles */
1540 sgen_init_pinning ();
1541 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1542 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1543 /* pin cemented objects */
1544 sgen_pin_cemented_objects ();
1545 /* identify pinned objects */
1546 sgen_optimize_pin_queue ();
1547 sgen_pinning_setup_section (nursery_section);
1549 pin_objects_in_nursery (FALSE, ctx);
1550 sgen_pinning_trim_queue_to_section (nursery_section);
1552 if (remset_consistency_checks)
1553 sgen_check_remset_consistency ();
1556 time_minor_pinning += TV_ELAPSED (btv, atv);
1557 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1558 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1560 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1561 sj->ops = object_ops;
1562 sj->gc_thread_gray_queue = &gc_thread_gray_queue;
1563 sgen_workers_enqueue_job (&sj->job, FALSE);
1565 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1567 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1568 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1570 sgen_pin_stats_report ();
1572 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1573 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1576 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1578 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1581 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1583 finish_gray_stack (GENERATION_NURSERY, ctx);
1586 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1587 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1589 if (objects_pinned) {
1590 sgen_optimize_pin_queue ();
1591 sgen_pinning_setup_section (nursery_section);
1595 * This is the latest point at which we can do this check, because
1596 * sgen_build_nursery_fragments() unpins nursery objects again.
1598 if (remset_consistency_checks)
1599 sgen_check_remset_consistency ();
1601 /* walk the pin_queue, build up the fragment list of free memory, unmark
1602 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1605 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1606 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1607 if (!fragment_total)
1610 /* Clear TLABs for all threads */
1611 sgen_clear_tlabs ();
1613 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1615 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1616 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1618 if (remset_consistency_checks)
1619 sgen_check_major_refs ();
1621 major_collector.finish_nursery_collection ();
1623 TV_GETTIME (last_minor_collection_end_tv);
1624 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1626 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1628 /* prepare the pin queue for the next collection */
1629 sgen_finish_pinning ();
1630 if (sgen_have_pending_finalizers ()) {
1631 SGEN_LOG (4, "Finalizer-thread wakeup");
1632 sgen_client_finalize_notify ();
1634 sgen_pin_stats_reset ();
1635 /* clear cemented hash */
1636 sgen_cement_clear_below_threshold ();
1638 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1640 remset.finish_minor_collection ();
1642 check_scan_starts ();
1644 binary_protocol_flush_buffers (FALSE);
1646 sgen_memgov_minor_collection_end (reason, is_overflow);
1648 /*objects are late pinned because of lack of memory, so a major is a good call*/
1649 needs_major = objects_pinned > 0;
1650 current_collection_generation = -1;
1653 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1655 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1656 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1662 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1663 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1664 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1665 } CopyOrMarkFromRootsMode;
1668 major_copy_or_mark_from_roots (SgenGrayQueue *gc_thread_gray_queue, size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1673 /* FIXME: only use these values for the precise scan
1674 * note that to_space pointers should be excluded anyway...
1676 char *heap_start = NULL;
1677 char *heap_end = (char*)-1;
1678 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, gc_thread_gray_queue);
1679 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1681 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1683 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1684 /*This cleans up unused fragments */
1685 sgen_nursery_allocator_prepare_for_pinning ();
1687 if (do_concurrent_checks)
1688 sgen_debug_check_nursery_is_clean ();
1690 /* The concurrent collector doesn't touch the nursery. */
1691 sgen_nursery_alloc_prepare_for_major ();
1696 /* Pinning depends on this */
1697 sgen_clear_nursery_fragments ();
1699 if (whole_heap_check_before_collection)
1700 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1703 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1705 if (!sgen_collection_is_concurrent ())
1706 nursery_section->next_data = sgen_get_nursery_end ();
1707 /* we should also coalesce scanning from sections close to each other
1708 * and deal with pointers outside of the sections later.
1713 sgen_client_pre_collection_checks ();
1715 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1716 /* Remsets are not useful for a major collection */
1717 remset.clear_cards ();
1720 sgen_process_fin_stage_entries ();
1723 sgen_init_pinning ();
1724 SGEN_LOG (6, "Collecting pinned addresses");
1725 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1726 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1727 /* Pin cemented objects that were forced */
1728 sgen_pin_cemented_objects ();
1730 sgen_optimize_pin_queue ();
1731 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1733 * Cemented objects that are in the pinned list will be marked. When
1734 * marking concurrently we won't mark mod-union cards for these objects.
1735 * Instead they will remain cemented until the next major collection,
1736 * when we will recheck if they are still pinned in the roots.
1738 sgen_cement_force_pinned ();
1741 sgen_client_collecting_major_1 ();
1744 * pin_queue now contains all candidate pointers, sorted and
1745 * uniqued. We must do two passes now to figure out which
1746 * objects are pinned.
1748 * The first is to find within the pin_queue the area for each
1749 * section. This requires that the pin_queue be sorted. We
1750 * also process the LOS objects and pinned chunks here.
1752 * The second, destructive, pass is to reduce the section
1753 * areas to pointers to the actually pinned objects.
1755 SGEN_LOG (6, "Pinning from sections");
1756 /* first pass for the sections */
1757 sgen_find_section_pin_queue_start_end (nursery_section);
1758 /* identify possible pointers to the insize of large objects */
1759 SGEN_LOG (6, "Pinning from large objects");
1760 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1762 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1763 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1765 if (sgen_los_object_is_pinned (bigobj->data)) {
1766 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1769 sgen_los_pin_object (bigobj->data);
1770 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1771 GRAY_OBJECT_ENQUEUE (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1772 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1773 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1774 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1775 (unsigned long)sgen_los_object_size (bigobj));
1777 sgen_client_pinned_los_object (bigobj->data);
1781 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1782 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1783 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1785 major_collector.pin_objects (gc_thread_gray_queue);
1786 if (old_next_pin_slot)
1787 *old_next_pin_slot = sgen_get_pinned_count ();
1790 * We don't actually pin when starting a concurrent collection, so the remset
1791 * consistency check won't work.
1793 if (remset_consistency_checks && mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1794 sgen_check_remset_consistency ();
1797 time_major_pinning += TV_ELAPSED (atv, btv);
1798 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1799 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1801 major_collector.init_to_space ();
1803 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1804 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1805 if (sgen_workers_have_idle_work ()) {
1807 * We force the finish of the worker with the new object ops context
1808 * which can also do copying. We need to have finished pinning.
1810 sgen_workers_start_all_workers (object_ops, NULL);
1811 sgen_workers_join ();
1815 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1816 main_gc_thread = mono_native_thread_self ();
1819 sgen_client_collecting_major_2 ();
1822 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1824 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1826 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops, FALSE);
1829 time_major_scan_roots += TV_ELAPSED (atv, btv);
1832 * We start the concurrent worker after pinning and after we scanned the roots
1833 * in order to make sure that the worker does not finish before handling all
1836 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1837 if (precleaning_enabled) {
1839 /* Mod union preclean job */
1840 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1841 sj->ops = object_ops;
1842 sj->gc_thread_gray_queue = NULL;
1843 sgen_workers_start_all_workers (object_ops, &sj->job);
1845 sgen_workers_start_all_workers (object_ops, NULL);
1847 gray_queue_enable_redirect (gc_thread_gray_queue);
1850 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1853 /* Mod union card table */
1854 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1855 sj->ops = object_ops;
1856 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1857 sgen_workers_enqueue_job (&sj->job, FALSE);
1859 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1860 sj->ops = object_ops;
1861 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1862 sgen_workers_enqueue_job (&sj->job, FALSE);
1865 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1868 sgen_pin_stats_report ();
1870 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1871 sgen_finish_pinning ();
1873 sgen_pin_stats_reset ();
1875 if (do_concurrent_checks)
1876 sgen_debug_check_nursery_is_clean ();
1881 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
1883 SgenObjectOperations *object_ops;
1885 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1887 current_collection_generation = GENERATION_OLD;
1889 sgen_workers_assert_gray_queue_is_empty ();
1892 sgen_cement_reset ();
1895 g_assert (major_collector.is_concurrent);
1896 concurrent_collection_in_progress = TRUE;
1898 object_ops = &major_collector.major_ops_concurrent_start;
1900 object_ops = &major_collector.major_ops_serial;
1903 reset_pinned_from_failed_allocation ();
1905 sgen_memgov_major_collection_start (concurrent, reason);
1907 //count_ref_nonref_objs ();
1908 //consistency_check ();
1910 check_scan_starts ();
1913 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1914 gc_stats.major_gc_count ++;
1916 if (major_collector.start_major_collection)
1917 major_collector.start_major_collection ();
1919 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);
1923 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
1925 ScannedObjectCounts counts;
1926 SgenObjectOperations *object_ops;
1927 mword fragment_total;
1933 if (concurrent_collection_in_progress) {
1934 object_ops = &major_collector.major_ops_concurrent_finish;
1936 major_copy_or_mark_from_roots (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1938 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1939 main_gc_thread = NULL;
1942 object_ops = &major_collector.major_ops_serial;
1945 sgen_workers_assert_gray_queue_is_empty ();
1947 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, gc_thread_gray_queue));
1949 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1951 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1953 if (objects_pinned) {
1954 g_assert (!concurrent_collection_in_progress);
1957 * This is slow, but we just OOM'd.
1959 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1960 * queue is laid out at this point.
1962 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1964 * We need to reestablish all pinned nursery objects in the pin queue
1965 * because they're needed for fragment creation. Unpinning happens by
1966 * walking the whole queue, so it's not necessary to reestablish where major
1967 * heap block pins are - all we care is that they're still in there
1970 sgen_optimize_pin_queue ();
1971 sgen_find_section_pin_queue_start_end (nursery_section);
1975 reset_heap_boundaries ();
1976 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1978 if (whole_heap_check_before_collection)
1979 sgen_check_whole_heap (FALSE);
1981 /* walk the pin_queue, build up the fragment list of free memory, unmark
1982 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1985 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1986 if (!fragment_total)
1988 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1990 if (do_concurrent_checks && concurrent_collection_in_progress)
1991 sgen_debug_check_nursery_is_clean ();
1993 /* prepare the pin queue for the next collection */
1994 sgen_finish_pinning ();
1996 /* Clear TLABs for all threads */
1997 sgen_clear_tlabs ();
1999 sgen_pin_stats_reset ();
2001 sgen_cement_clear_below_threshold ();
2003 if (check_mark_bits_after_major_collection)
2004 sgen_check_heap_marked (concurrent_collection_in_progress);
2007 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2009 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2010 sgen_memgov_major_pre_sweep ();
2013 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2018 time_major_los_sweep += TV_ELAPSED (atv, btv);
2020 major_collector.sweep ();
2022 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2025 time_major_sweep += TV_ELAPSED (btv, atv);
2027 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2029 if (sgen_have_pending_finalizers ()) {
2030 SGEN_LOG (4, "Finalizer-thread wakeup");
2031 sgen_client_finalize_notify ();
2034 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2035 current_collection_generation = -1;
2037 memset (&counts, 0, sizeof (ScannedObjectCounts));
2038 major_collector.finish_major_collection (&counts);
2040 sgen_workers_assert_gray_queue_is_empty ();
2042 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2043 if (concurrent_collection_in_progress)
2044 concurrent_collection_in_progress = FALSE;
2046 check_scan_starts ();
2048 binary_protocol_flush_buffers (FALSE);
2050 //consistency_check ();
2052 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2056 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2058 TV_DECLARE (time_start);
2059 TV_DECLARE (time_end);
2060 size_t old_next_pin_slot;
2061 SgenGrayQueue gc_thread_gray_queue;
2063 if (disable_major_collections)
2066 if (major_collector.get_and_reset_num_major_objects_marked) {
2067 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2068 g_assert (!num_marked);
2071 /* world must be stopped already */
2072 TV_GETTIME (time_start);
2074 init_gray_queue (&gc_thread_gray_queue, FALSE);
2075 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2076 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2077 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2079 TV_GETTIME (time_end);
2080 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2082 /* FIXME: also report this to the user, preferably in gc-end. */
2083 if (major_collector.get_and_reset_num_major_objects_marked)
2084 major_collector.get_and_reset_num_major_objects_marked ();
2086 return bytes_pinned_from_failed_allocation > 0;
2090 major_start_concurrent_collection (const char *reason)
2092 TV_DECLARE (time_start);
2093 TV_DECLARE (time_end);
2094 long long num_objects_marked;
2095 SgenGrayQueue gc_thread_gray_queue;
2097 if (disable_major_collections)
2100 TV_GETTIME (time_start);
2101 SGEN_TV_GETTIME (time_major_conc_collection_start);
2103 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2104 g_assert (num_objects_marked == 0);
2106 binary_protocol_concurrent_start ();
2108 init_gray_queue (&gc_thread_gray_queue, TRUE);
2109 // FIXME: store reason and pass it when finishing
2110 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2111 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2113 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2115 TV_GETTIME (time_end);
2116 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2118 current_collection_generation = -1;
2122 * Returns whether the major collection has finished.
2125 major_should_finish_concurrent_collection (void)
2127 return sgen_workers_all_done ();
2131 major_update_concurrent_collection (void)
2133 TV_DECLARE (total_start);
2134 TV_DECLARE (total_end);
2136 TV_GETTIME (total_start);
2138 binary_protocol_concurrent_update ();
2140 major_collector.update_cardtable_mod_union ();
2141 sgen_los_update_cardtable_mod_union ();
2143 TV_GETTIME (total_end);
2144 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2148 major_finish_concurrent_collection (gboolean forced)
2150 SgenGrayQueue gc_thread_gray_queue;
2151 TV_DECLARE (total_start);
2152 TV_DECLARE (total_end);
2154 TV_GETTIME (total_start);
2156 binary_protocol_concurrent_finish ();
2159 * We need to stop all workers since we're updating the cardtable below.
2160 * The workers will be resumed with a finishing pause context to avoid
2161 * additional cardtable and object scanning.
2163 sgen_workers_stop_all_workers ();
2165 SGEN_TV_GETTIME (time_major_conc_collection_end);
2166 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2168 major_collector.update_cardtable_mod_union ();
2169 sgen_los_update_cardtable_mod_union ();
2171 if (mod_union_consistency_check)
2172 sgen_check_mod_union_consistency ();
2174 current_collection_generation = GENERATION_OLD;
2175 sgen_cement_reset ();
2176 init_gray_queue (&gc_thread_gray_queue, FALSE);
2177 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2178 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2180 TV_GETTIME (total_end);
2181 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2183 current_collection_generation = -1;
2187 * Ensure an allocation request for @size will succeed by freeing enough memory.
2189 * LOCKING: The GC lock MUST be held.
2192 sgen_ensure_free_space (size_t size, int generation)
2194 int generation_to_collect = -1;
2195 const char *reason = NULL;
2197 if (generation == GENERATION_OLD) {
2198 if (sgen_need_major_collection (size)) {
2199 reason = "LOS overflow";
2200 generation_to_collect = GENERATION_OLD;
2203 if (degraded_mode) {
2204 if (sgen_need_major_collection (size)) {
2205 reason = "Degraded mode overflow";
2206 generation_to_collect = GENERATION_OLD;
2208 } else if (sgen_need_major_collection (size)) {
2209 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2210 generation_to_collect = GENERATION_OLD;
2212 generation_to_collect = GENERATION_NURSERY;
2213 reason = "Nursery full";
2217 if (generation_to_collect == -1) {
2218 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2219 generation_to_collect = GENERATION_OLD;
2220 reason = "Finish concurrent collection";
2224 if (generation_to_collect == -1)
2226 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2230 * LOCKING: Assumes the GC lock is held.
2233 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2235 TV_DECLARE (gc_total_start);
2236 TV_DECLARE (gc_total_end);
2237 int overflow_generation_to_collect = -1;
2238 int oldest_generation_collected = generation_to_collect;
2239 const char *overflow_reason = NULL;
2240 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2242 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2244 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2247 sgen_stop_world (generation_to_collect);
2249 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2252 TV_GETTIME (gc_total_start);
2254 // FIXME: extract overflow reason
2255 // FIXME: minor overflow for concurrent case
2256 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2257 if (concurrent_collection_in_progress)
2258 major_update_concurrent_collection ();
2260 if (collect_nursery (reason, FALSE, NULL, FALSE) && !concurrent_collection_in_progress) {
2261 overflow_generation_to_collect = GENERATION_OLD;
2262 overflow_reason = "Minor overflow";
2264 } else if (finish_concurrent) {
2265 major_finish_concurrent_collection (wait_to_finish);
2266 oldest_generation_collected = GENERATION_OLD;
2268 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2269 if (major_collector.is_concurrent && !wait_to_finish) {
2270 collect_nursery ("Concurrent start", FALSE, NULL, FALSE);
2271 major_start_concurrent_collection (reason);
2272 oldest_generation_collected = GENERATION_NURSERY;
2273 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2274 overflow_generation_to_collect = GENERATION_NURSERY;
2275 overflow_reason = "Excessive pinning";
2279 if (overflow_generation_to_collect != -1) {
2280 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2283 * We need to do an overflow collection, either because we ran out of memory
2284 * or the nursery is fully pinned.
2287 if (overflow_generation_to_collect == GENERATION_NURSERY)
2288 collect_nursery (overflow_reason, TRUE, NULL, FALSE);
2290 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2292 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2295 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2297 /* this also sets the proper pointers for the next allocation */
2298 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2299 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2300 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2301 sgen_dump_pin_queue ();
2305 TV_GETTIME (gc_total_end);
2306 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2309 sgen_restart_world (oldest_generation_collected);
2313 * ######################################################################
2314 * ######## Memory allocation from the OS
2315 * ######################################################################
2316 * This section of code deals with getting memory from the OS and
2317 * allocating memory for GC-internal data structures.
2318 * Internal memory can be handled with a freelist for small objects.
2324 G_GNUC_UNUSED static void
2325 report_internal_mem_usage (void)
2327 printf ("Internal memory usage:\n");
2328 sgen_report_internal_mem_usage ();
2329 printf ("Pinned memory usage:\n");
2330 major_collector.report_pinned_memory_usage ();
2334 * ######################################################################
2335 * ######## Finalization support
2336 * ######################################################################
2340 * If the object has been forwarded it means it's still referenced from a root.
2341 * If it is pinned it's still alive as well.
2342 * A LOS object is only alive if we have pinned it.
2343 * Return TRUE if @obj is ready to be finalized.
2345 static inline gboolean
2346 sgen_is_object_alive (GCObject *object)
2348 if (ptr_in_nursery (object))
2349 return sgen_nursery_is_object_alive (object);
2351 return sgen_major_is_object_alive (object);
2355 * This function returns true if @object is either alive and belongs to the
2356 * current collection - major collections are full heap, so old gen objects
2357 * are never alive during a minor collection.
2360 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2362 if (ptr_in_nursery (object))
2363 return sgen_nursery_is_object_alive (object);
2365 if (current_collection_generation == GENERATION_NURSERY)
2368 return sgen_major_is_object_alive (object);
2373 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2375 return !sgen_is_object_alive (object);
2379 sgen_queue_finalization_entry (GCObject *obj)
2381 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2383 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2385 sgen_client_object_queued_for_finalization (obj);
2389 sgen_object_is_live (GCObject *obj)
2391 return sgen_is_object_alive_and_on_current_collection (obj);
2395 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2396 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2397 * all finalizers have really finished running.
2399 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2400 * This means that just checking whether the queues are empty leaves the possibility that an
2401 * object might have been dequeued but not yet finalized. That's why we need the additional
2402 * flag `pending_unqueued_finalizer`.
2405 static volatile gboolean pending_unqueued_finalizer = FALSE;
2406 volatile gboolean sgen_suspend_finalizers = FALSE;
2409 sgen_set_suspend_finalizers (void)
2411 sgen_suspend_finalizers = TRUE;
2415 sgen_gc_invoke_finalizers (void)
2419 g_assert (!pending_unqueued_finalizer);
2421 /* FIXME: batch to reduce lock contention */
2422 while (sgen_have_pending_finalizers ()) {
2428 * We need to set `pending_unqueued_finalizer` before dequeing the
2429 * finalizable object.
2431 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2432 pending_unqueued_finalizer = TRUE;
2433 mono_memory_write_barrier ();
2434 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2435 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2436 pending_unqueued_finalizer = TRUE;
2437 mono_memory_write_barrier ();
2438 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2444 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2452 /* the object is on the stack so it is pinned */
2453 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2454 sgen_client_run_finalize (obj);
2457 if (pending_unqueued_finalizer) {
2458 mono_memory_write_barrier ();
2459 pending_unqueued_finalizer = FALSE;
2466 sgen_have_pending_finalizers (void)
2468 if (sgen_suspend_finalizers)
2470 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2474 * ######################################################################
2475 * ######## registered roots support
2476 * ######################################################################
2480 * We do not coalesce roots.
2483 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2485 RootRecord new_root;
2488 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2489 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2490 /* we allow changing the size and the descriptor (for thread statics etc) */
2492 size_t old_size = root->end_root - start;
2493 root->end_root = start + size;
2494 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2495 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2496 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2497 root->root_desc = descr;
2499 roots_size -= old_size;
2505 new_root.end_root = start + size;
2506 new_root.root_desc = descr;
2507 new_root.source = source;
2510 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2513 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);
2520 sgen_deregister_root (char* addr)
2526 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2527 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2528 roots_size -= (root.end_root - addr);
2534 * ######################################################################
2535 * ######## Thread handling (stop/start code)
2536 * ######################################################################
2540 sgen_get_current_collection_generation (void)
2542 return current_collection_generation;
2546 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2548 #ifndef HAVE_KW_THREAD
2549 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2552 sgen_init_tlab_info (info);
2554 sgen_client_thread_register (info, stack_bottom_fallback);
2560 sgen_thread_unregister (SgenThreadInfo *p)
2562 sgen_client_thread_unregister (p);
2566 * ######################################################################
2567 * ######## Write barriers
2568 * ######################################################################
2572 * Note: the write barriers first do the needed GC work and then do the actual store:
2573 * this way the value is visible to the conservative GC scan after the write barrier
2574 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2575 * the conservative scan, otherwise by the remembered set scan.
2579 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2581 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2582 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2583 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2584 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2588 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2589 if (binary_protocol_is_heavy_enabled ()) {
2591 for (i = 0; i < count; ++i) {
2592 gpointer dest = (gpointer*)dest_ptr + i;
2593 gpointer obj = *((gpointer*)src_ptr + i);
2595 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2600 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2604 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2608 HEAVY_STAT (++stat_wbarrier_generic_store);
2610 sgen_client_wbarrier_generic_nostore_check (ptr);
2612 obj = *(gpointer*)ptr;
2614 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2617 * We need to record old->old pointer locations for the
2618 * concurrent collector.
2620 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2621 SGEN_LOG (8, "Skipping remset at %p", ptr);
2625 SGEN_LOG (8, "Adding remset at %p", ptr);
2627 remset.wbarrier_generic_nostore (ptr);
2631 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2633 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2634 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2635 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2636 mono_gc_wbarrier_generic_nostore (ptr);
2637 sgen_dummy_use (value);
2640 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2641 * as an atomic operation with release semantics.
2644 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2646 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2648 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2650 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2652 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2653 mono_gc_wbarrier_generic_nostore (ptr);
2655 sgen_dummy_use (value);
2659 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2661 GCObject **dest = (GCObject **)_dest;
2662 GCObject **src = (GCObject **)_src;
2666 mono_gc_wbarrier_generic_store (dest, *src);
2671 size -= SIZEOF_VOID_P;
2677 * ######################################################################
2678 * ######## Other mono public interface functions.
2679 * ######################################################################
2683 sgen_gc_collect (int generation)
2688 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2693 sgen_gc_collection_count (int generation)
2695 if (generation == 0)
2696 return gc_stats.minor_gc_count;
2697 return gc_stats.major_gc_count;
2701 sgen_gc_get_used_size (void)
2705 tot = los_memory_usage;
2706 tot += nursery_section->next_data - nursery_section->data;
2707 tot += major_collector.get_used_size ();
2708 /* FIXME: account for pinned objects */
2714 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2718 va_start (ap, description_format);
2720 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2721 vfprintf (stderr, description_format, ap);
2723 fprintf (stderr, " - %s", fallback);
2724 fprintf (stderr, "\n");
2730 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2733 double val = strtod (opt, &endptr);
2734 if (endptr == opt) {
2735 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2738 else if (val < min || val > max) {
2739 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2751 char *major_collector_opt = NULL;
2752 char *minor_collector_opt = NULL;
2753 size_t max_heap = 0;
2754 size_t soft_limit = 0;
2756 gboolean debug_print_allowance = FALSE;
2757 double allowance_ratio = 0, save_target = 0;
2758 gboolean cement_enabled = TRUE;
2761 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2764 /* already inited */
2767 /* being inited by another thread */
2768 mono_thread_info_usleep (1000);
2771 /* we will init it */
2774 g_assert_not_reached ();
2776 } while (result != 0);
2778 SGEN_TV_GETTIME (sgen_init_timestamp);
2780 #ifdef SGEN_WITHOUT_MONO
2781 mono_thread_smr_init ();
2784 mono_coop_mutex_init (&gc_mutex);
2786 gc_debug_file = stderr;
2788 mono_coop_mutex_init (&sgen_interruption_mutex);
2790 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2791 opts = g_strsplit (env, ",", -1);
2792 for (ptr = opts; *ptr; ++ptr) {
2794 if (g_str_has_prefix (opt, "major=")) {
2795 opt = strchr (opt, '=') + 1;
2796 major_collector_opt = g_strdup (opt);
2797 } else if (g_str_has_prefix (opt, "minor=")) {
2798 opt = strchr (opt, '=') + 1;
2799 minor_collector_opt = g_strdup (opt);
2807 sgen_init_internal_allocator ();
2808 sgen_init_nursery_allocator ();
2809 sgen_init_fin_weak_hash ();
2810 sgen_init_hash_table ();
2811 sgen_init_descriptors ();
2812 sgen_init_gray_queues ();
2813 sgen_init_allocator ();
2814 sgen_init_gchandles ();
2816 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2817 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2819 sgen_client_init ();
2821 if (!minor_collector_opt) {
2822 sgen_simple_nursery_init (&sgen_minor_collector);
2824 if (!strcmp (minor_collector_opt, "simple")) {
2826 sgen_simple_nursery_init (&sgen_minor_collector);
2827 } else if (!strcmp (minor_collector_opt, "split")) {
2828 sgen_split_nursery_init (&sgen_minor_collector);
2830 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2831 goto use_simple_nursery;
2835 if (!major_collector_opt) {
2837 DEFAULT_MAJOR_INIT (&major_collector);
2838 } else if (!strcmp (major_collector_opt, "marksweep")) {
2839 sgen_marksweep_init (&major_collector);
2840 } else if (!strcmp (major_collector_opt, "marksweep-conc")) {
2841 sgen_marksweep_conc_init (&major_collector);
2843 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `" DEFAULT_MAJOR_NAME "` instead.", "Unknown major collector `%s'.", major_collector_opt);
2844 goto use_default_major;
2847 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2850 gboolean usage_printed = FALSE;
2852 for (ptr = opts; *ptr; ++ptr) {
2854 if (!strcmp (opt, ""))
2856 if (g_str_has_prefix (opt, "major="))
2858 if (g_str_has_prefix (opt, "minor="))
2860 if (g_str_has_prefix (opt, "max-heap-size=")) {
2861 size_t page_size = mono_pagesize ();
2862 size_t max_heap_candidate = 0;
2863 opt = strchr (opt, '=') + 1;
2864 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2865 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2866 if (max_heap != max_heap_candidate)
2867 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2869 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2873 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2874 opt = strchr (opt, '=') + 1;
2875 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2876 if (soft_limit <= 0) {
2877 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2881 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2887 if (g_str_has_prefix (opt, "nursery-size=")) {
2889 opt = strchr (opt, '=') + 1;
2890 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2891 if ((val & (val - 1))) {
2892 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2896 if (val < SGEN_MAX_NURSERY_WASTE) {
2897 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2898 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2902 sgen_nursery_size = val;
2903 sgen_nursery_bits = 0;
2904 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2907 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2913 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2915 opt = strchr (opt, '=') + 1;
2916 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2917 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2922 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2924 opt = strchr (opt, '=') + 1;
2925 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2926 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2927 allowance_ratio = val;
2932 if (!strcmp (opt, "cementing")) {
2933 cement_enabled = TRUE;
2936 if (!strcmp (opt, "no-cementing")) {
2937 cement_enabled = FALSE;
2941 if (!strcmp (opt, "precleaning")) {
2942 precleaning_enabled = TRUE;
2945 if (!strcmp (opt, "no-precleaning")) {
2946 precleaning_enabled = FALSE;
2950 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2953 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2956 if (sgen_client_handle_gc_param (opt))
2959 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2964 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2965 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2966 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2967 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2968 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2969 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2970 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2971 fprintf (stderr, " [no-]cementing\n");
2972 if (major_collector.print_gc_param_usage)
2973 major_collector.print_gc_param_usage ();
2974 if (sgen_minor_collector.print_gc_param_usage)
2975 sgen_minor_collector.print_gc_param_usage ();
2976 sgen_client_print_gc_params_usage ();
2977 fprintf (stderr, " Experimental options:\n");
2978 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2979 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);
2980 fprintf (stderr, "\n");
2982 usage_printed = TRUE;
2987 if (major_collector_opt)
2988 g_free (major_collector_opt);
2990 if (minor_collector_opt)
2991 g_free (minor_collector_opt);
2995 sgen_pinning_init ();
2996 sgen_cement_init (cement_enabled);
2998 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
2999 gboolean usage_printed = FALSE;
3001 opts = g_strsplit (env, ",", -1);
3002 for (ptr = opts; ptr && *ptr; ptr ++) {
3004 if (!strcmp (opt, ""))
3006 if (opt [0] >= '0' && opt [0] <= '9') {
3007 gc_debug_level = atoi (opt);
3012 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3013 gc_debug_file = fopen (rf, "wb");
3015 gc_debug_file = stderr;
3018 } else if (!strcmp (opt, "print-allowance")) {
3019 debug_print_allowance = TRUE;
3020 } else if (!strcmp (opt, "print-pinning")) {
3021 sgen_pin_stats_enable ();
3022 } else if (!strcmp (opt, "verify-before-allocs")) {
3023 verify_before_allocs = 1;
3024 has_per_allocation_action = TRUE;
3025 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3026 char *arg = strchr (opt, '=') + 1;
3027 verify_before_allocs = atoi (arg);
3028 has_per_allocation_action = TRUE;
3029 } else if (!strcmp (opt, "collect-before-allocs")) {
3030 collect_before_allocs = 1;
3031 has_per_allocation_action = TRUE;
3032 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3033 char *arg = strchr (opt, '=') + 1;
3034 has_per_allocation_action = TRUE;
3035 collect_before_allocs = atoi (arg);
3036 } else if (!strcmp (opt, "verify-before-collections")) {
3037 whole_heap_check_before_collection = TRUE;
3038 } else if (!strcmp (opt, "check-remset-consistency")) {
3039 remset_consistency_checks = TRUE;
3040 nursery_clear_policy = CLEAR_AT_GC;
3041 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3042 if (!major_collector.is_concurrent) {
3043 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3046 mod_union_consistency_check = TRUE;
3047 } else if (!strcmp (opt, "check-mark-bits")) {
3048 check_mark_bits_after_major_collection = TRUE;
3049 } else if (!strcmp (opt, "check-nursery-pinned")) {
3050 check_nursery_objects_pinned = TRUE;
3051 } else if (!strcmp (opt, "clear-at-gc")) {
3052 nursery_clear_policy = CLEAR_AT_GC;
3053 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3054 nursery_clear_policy = CLEAR_AT_GC;
3055 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3056 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3057 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3058 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3059 } else if (!strcmp (opt, "check-scan-starts")) {
3060 do_scan_starts_check = TRUE;
3061 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3062 do_verify_nursery = TRUE;
3063 } else if (!strcmp (opt, "check-concurrent")) {
3064 if (!major_collector.is_concurrent) {
3065 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3068 nursery_clear_policy = CLEAR_AT_GC;
3069 do_concurrent_checks = TRUE;
3070 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3071 do_dump_nursery_content = TRUE;
3072 } else if (!strcmp (opt, "disable-minor")) {
3073 disable_minor_collections = TRUE;
3074 } else if (!strcmp (opt, "disable-major")) {
3075 disable_major_collections = TRUE;
3076 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3077 char *filename = strchr (opt, '=') + 1;
3078 nursery_clear_policy = CLEAR_AT_GC;
3079 sgen_debug_enable_heap_dump (filename);
3080 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3081 char *filename = strchr (opt, '=') + 1;
3082 char *colon = strrchr (filename, ':');
3085 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3086 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3091 binary_protocol_init (filename, (long long)limit);
3092 } else if (!strcmp (opt, "nursery-canaries")) {
3093 do_verify_nursery = TRUE;
3094 enable_nursery_canaries = TRUE;
3095 } else if (!sgen_client_handle_gc_debug (opt)) {
3096 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3101 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);
3102 fprintf (stderr, "Valid <option>s are:\n");
3103 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3104 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3105 fprintf (stderr, " check-remset-consistency\n");
3106 fprintf (stderr, " check-mark-bits\n");
3107 fprintf (stderr, " check-nursery-pinned\n");
3108 fprintf (stderr, " verify-before-collections\n");
3109 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3110 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3111 fprintf (stderr, " disable-minor\n");
3112 fprintf (stderr, " disable-major\n");
3113 fprintf (stderr, " check-concurrent\n");
3114 fprintf (stderr, " clear-[nursery-]at-gc\n");
3115 fprintf (stderr, " clear-at-tlab-creation\n");
3116 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3117 fprintf (stderr, " check-scan-starts\n");
3118 fprintf (stderr, " print-allowance\n");
3119 fprintf (stderr, " print-pinning\n");
3120 fprintf (stderr, " heap-dump=<filename>\n");
3121 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3122 fprintf (stderr, " nursery-canaries\n");
3123 sgen_client_print_gc_debug_usage ();
3124 fprintf (stderr, "\n");
3126 usage_printed = TRUE;
3132 if (check_mark_bits_after_major_collection)
3133 nursery_clear_policy = CLEAR_AT_GC;
3135 if (major_collector.post_param_init)
3136 major_collector.post_param_init (&major_collector);
3138 if (major_collector.needs_thread_pool)
3139 sgen_workers_init (1);
3141 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3143 memset (&remset, 0, sizeof (remset));
3145 sgen_card_table_init (&remset);
3147 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");
3151 sgen_init_bridge ();
3155 sgen_gc_initialized ()
3157 return gc_initialized > 0;
3161 sgen_get_nursery_clear_policy (void)
3163 return nursery_clear_policy;
3169 mono_coop_mutex_lock (&gc_mutex);
3173 sgen_gc_unlock (void)
3175 mono_coop_mutex_unlock (&gc_mutex);
3179 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3181 major_collector.iterate_live_block_ranges (callback);
3185 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3187 major_collector.iterate_block_ranges (callback);
3191 sgen_get_major_collector (void)
3193 return &major_collector;
3197 sgen_get_remset (void)
3203 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3205 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3206 sgen_los_count_cards (los_total, los_marked);
3209 static gboolean world_is_stopped = FALSE;
3211 /* LOCKING: assumes the GC lock is held */
3213 sgen_stop_world (int generation)
3215 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3217 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3219 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3221 sgen_client_stop_world (generation);
3223 world_is_stopped = TRUE;
3225 if (binary_protocol_is_heavy_enabled ())
3226 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3227 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3230 /* LOCKING: assumes the GC lock is held */
3232 sgen_restart_world (int generation)
3234 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3237 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3239 if (binary_protocol_is_heavy_enabled ())
3240 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3241 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3243 world_is_stopped = FALSE;
3245 sgen_client_restart_world (generation, &stw_time);
3247 binary_protocol_world_restarted (generation, sgen_timestamp ());
3249 if (sgen_client_bridge_need_processing ())
3250 sgen_client_bridge_processing_finish (generation);
3252 sgen_memgov_collection_end (generation, stw_time);
3256 sgen_is_world_stopped (void)
3258 return world_is_stopped;
3262 sgen_check_whole_heap_stw (void)
3264 sgen_stop_world (0);
3265 sgen_clear_nursery_fragments ();
3266 sgen_check_whole_heap (FALSE);
3267 sgen_restart_world (0);
3271 sgen_timestamp (void)
3273 SGEN_TV_DECLARE (timestamp);
3274 SGEN_TV_GETTIME (timestamp);
3275 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3278 #endif /* HAVE_SGEN_GC */