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 heap consistency check before each minor collection */
218 static gboolean consistency_check_at_minor_collection = 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 static SGEN_TV_DECLARE (last_minor_collection_start_tv);
292 static SGEN_TV_DECLARE (last_minor_collection_end_tv);
294 int gc_debug_level = 0;
299 mono_gc_flush_info (void)
301 fflush (gc_debug_file);
305 #define TV_DECLARE SGEN_TV_DECLARE
306 #define TV_GETTIME SGEN_TV_GETTIME
307 #define TV_ELAPSED SGEN_TV_ELAPSED
309 static SGEN_TV_DECLARE (sgen_init_timestamp);
311 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
313 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
314 #define object_is_pinned SGEN_OBJECT_IS_PINNED
315 #define pin_object SGEN_PIN_OBJECT
317 #define ptr_in_nursery sgen_ptr_in_nursery
319 #define LOAD_VTABLE SGEN_LOAD_VTABLE
322 nursery_canaries_enabled (void)
324 return enable_nursery_canaries;
327 #define safe_object_get_size sgen_safe_object_get_size
330 * ######################################################################
331 * ######## Global data.
332 * ######################################################################
334 MonoCoopMutex gc_mutex;
336 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
338 size_t degraded_mode = 0;
340 static mword bytes_pinned_from_failed_allocation = 0;
342 GCMemSection *nursery_section = NULL;
343 static volatile mword lowest_heap_address = ~(mword)0;
344 static volatile mword highest_heap_address = 0;
346 MonoCoopMutex sgen_interruption_mutex;
348 int current_collection_generation = -1;
349 static volatile gboolean concurrent_collection_in_progress = FALSE;
351 /* objects that are ready to be finalized */
352 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
353 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
355 /* registered roots: the key to the hash is the root start address */
357 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
359 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
360 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
361 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
362 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
364 static mword roots_size = 0; /* amount of memory in the root set */
366 /* The size of a TLAB */
367 /* The bigger the value, the less often we have to go to the slow path to allocate a new
368 * one, but the more space is wasted by threads not allocating much memory.
370 * FIXME: Make this self-tuning for each thread.
372 guint32 tlab_size = (1024 * 4);
374 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
376 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
378 #define ALIGN_UP SGEN_ALIGN_UP
380 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
381 MonoNativeThreadId main_gc_thread = NULL;
384 /*Object was pinned during the current collection*/
385 static mword objects_pinned;
388 * ######################################################################
389 * ######## Macros and function declarations.
390 * ######################################################################
393 typedef SgenGrayQueue GrayQueue;
395 /* forward declarations */
396 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
398 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
399 static void finish_gray_stack (int generation, ScanCopyContext ctx);
402 SgenMajorCollector major_collector;
403 SgenMinorCollector sgen_minor_collector;
404 /* FIXME: get rid of this */
405 static GrayQueue gray_queue;
407 static SgenRememberedSet remset;
409 /* The gray queue to use from the main collection thread. */
410 #define WORKERS_DISTRIBUTE_GRAY_QUEUE (&gray_queue)
413 * The gray queue a worker job must use. If we're not parallel or
414 * concurrent, we use the main gray queue.
416 static SgenGrayQueue*
417 sgen_workers_get_job_gray_queue (WorkerData *worker_data)
419 return worker_data ? &worker_data->private_gray_queue : WORKERS_DISTRIBUTE_GRAY_QUEUE;
423 gray_queue_redirect (SgenGrayQueue *queue)
425 gboolean wake = FALSE;
428 GrayQueueSection *section = sgen_gray_object_dequeue_section (queue);
431 sgen_section_gray_queue_enqueue ((SgenSectionGrayQueue *)queue->alloc_prepare_data, section);
436 g_assert (concurrent_collection_in_progress);
437 sgen_workers_ensure_awake ();
442 gray_queue_enable_redirect (SgenGrayQueue *queue)
444 if (!concurrent_collection_in_progress)
447 sgen_gray_queue_set_alloc_prepare (queue, gray_queue_redirect, sgen_workers_get_distribute_section_gray_queue ());
448 gray_queue_redirect (queue);
452 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
454 while (start < end) {
458 if (!*(void**)start) {
459 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
464 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
470 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
471 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
472 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
473 callback ((GCObject*)obj, size, data);
474 CANARIFY_SIZE (size);
476 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
484 * sgen_add_to_global_remset:
486 * The global remset contains locations which point into newspace after
487 * a minor collection. This can happen if the objects they point to are pinned.
489 * LOCKING: If called from a parallel collector, the global remset
490 * lock must be held. For serial collectors that is not necessary.
493 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
495 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
497 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
499 if (!major_collector.is_concurrent) {
500 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
502 if (current_collection_generation == -1)
503 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
506 if (!object_is_pinned (obj))
507 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");
508 else if (sgen_cement_lookup_or_register (obj))
511 remset.record_pointer (ptr);
513 sgen_pin_stats_register_global_remset (obj);
515 SGEN_LOG (8, "Adding global remset for %p", ptr);
516 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
520 * sgen_drain_gray_stack:
522 * Scan objects in the gray stack until the stack is empty. This should be called
523 * frequently after each object is copied, to achieve better locality and cache
528 sgen_drain_gray_stack (ScanCopyContext ctx)
530 ScanObjectFunc scan_func = ctx.ops->scan_object;
531 GrayQueue *queue = ctx.queue;
533 if (ctx.ops->drain_gray_stack)
534 return ctx.ops->drain_gray_stack (queue);
539 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
542 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
543 scan_func (obj, desc, queue);
549 * Addresses in the pin queue are already sorted. This function finds
550 * the object header for each address and pins the object. The
551 * addresses must be inside the nursery section. The (start of the)
552 * address array is overwritten with the addresses of the actually
553 * pinned objects. Return the number of pinned objects.
556 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
558 GCMemSection *section = nursery_section;
559 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
560 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
561 void *start_nursery = section->data;
562 void *end_nursery = section->next_data;
567 void *pinning_front = start_nursery;
569 void **definitely_pinned = start;
570 ScanObjectFunc scan_func = ctx.ops->scan_object;
571 SgenGrayQueue *queue = ctx.queue;
573 sgen_nursery_allocator_prepare_for_pinning ();
575 while (start < end) {
576 GCObject *obj_to_pin = NULL;
577 size_t obj_to_pin_size = 0;
582 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
583 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
590 SGEN_LOG (5, "Considering pinning addr %p", addr);
591 /* We've already processed everything up to pinning_front. */
592 if (addr < pinning_front) {
598 * Find the closest scan start <= addr. We might search backward in the
599 * scan_starts array because entries might be NULL. In the worst case we
600 * start at start_nursery.
602 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
603 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
604 search_start = (void*)section->scan_starts [idx];
605 if (!search_start || search_start > addr) {
608 search_start = section->scan_starts [idx];
609 if (search_start && search_start <= addr)
612 if (!search_start || search_start > addr)
613 search_start = start_nursery;
617 * If the pinning front is closer than the scan start we found, start
618 * searching at the front.
620 if (search_start < pinning_front)
621 search_start = pinning_front;
624 * Now addr should be in an object a short distance from search_start.
626 * search_start must point to zeroed mem or point to an object.
629 size_t obj_size, canarified_obj_size;
632 if (!*(void**)search_start) {
633 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
634 /* The loop condition makes sure we don't overrun addr. */
638 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
641 * Filler arrays are marked by an invalid sync word. We don't
642 * consider them for pinning. They are not delimited by canaries,
645 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
646 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
647 CANARIFY_SIZE (canarified_obj_size);
649 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
650 /* This is the object we're looking for. */
651 obj_to_pin = (GCObject*)search_start;
652 obj_to_pin_size = canarified_obj_size;
657 /* Skip to the next object */
658 search_start = (void*)((char*)search_start + canarified_obj_size);
659 } while (search_start <= addr);
661 /* We've searched past the address we were looking for. */
663 pinning_front = search_start;
664 goto next_pin_queue_entry;
668 * We've found an object to pin. It might still be a dummy array, but we
669 * can advance the pinning front in any case.
671 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
674 * If this is a dummy array marking the beginning of a nursery
675 * fragment, we don't pin it.
677 if (sgen_client_object_is_array_fill (obj_to_pin))
678 goto next_pin_queue_entry;
681 * Finally - pin the object!
683 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
684 if (do_scan_objects) {
685 scan_func (obj_to_pin, desc, queue);
687 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
688 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
689 binary_protocol_pin (obj_to_pin,
690 (gpointer)LOAD_VTABLE (obj_to_pin),
691 safe_object_get_size (obj_to_pin));
693 pin_object (obj_to_pin);
694 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
695 sgen_pin_stats_register_object (obj_to_pin, obj_to_pin_size);
696 definitely_pinned [count] = obj_to_pin;
700 next_pin_queue_entry:
704 sgen_client_nursery_objects_pinned (definitely_pinned, count);
705 stat_pinned_objects += count;
710 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
714 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
717 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
718 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
722 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
723 * when we can't promote an object because we're out of memory.
726 sgen_pin_object (GCObject *object, GrayQueue *queue)
729 * All pinned objects are assumed to have been staged, so we need to stage as well.
730 * Also, the count of staged objects shows that "late pinning" happened.
732 sgen_pin_stage_ptr (object);
734 SGEN_PIN_OBJECT (object);
735 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
738 sgen_pin_stats_register_object (object, safe_object_get_size (object));
740 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
743 /* Sort the addresses in array in increasing order.
744 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
747 sgen_sort_addresses (void **array, size_t size)
752 for (i = 1; i < size; ++i) {
755 size_t parent = (child - 1) / 2;
757 if (array [parent] >= array [child])
760 tmp = array [parent];
761 array [parent] = array [child];
768 for (i = size - 1; i > 0; --i) {
771 array [i] = array [0];
777 while (root * 2 + 1 <= end) {
778 size_t child = root * 2 + 1;
780 if (child < end && array [child] < array [child + 1])
782 if (array [root] >= array [child])
786 array [root] = array [child];
795 * Scan the memory between start and end and queue values which could be pointers
796 * to the area between start_nursery and end_nursery for later consideration.
797 * Typically used for thread stacks.
800 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
804 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
806 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
807 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
810 while (start < end) {
812 * *start can point to the middle of an object
813 * note: should we handle pointing at the end of an object?
814 * pinning in C# code disallows pointing at the end of an object
815 * but there is some small chance that an optimizing C compiler
816 * may keep the only reference to an object by pointing
817 * at the end of it. We ignore this small chance for now.
818 * Pointers to the end of an object are indistinguishable
819 * from pointers to the start of the next object in memory
820 * so if we allow that we'd need to pin two objects...
821 * We queue the pointer in an array, the
822 * array will then be sorted and uniqued. This way
823 * we can coalesce several pinning pointers and it should
824 * be faster since we'd do a memory scan with increasing
825 * addresses. Note: we can align the address to the allocation
826 * alignment, so the unique process is more effective.
828 mword addr = (mword)*start;
829 addr &= ~(ALLOC_ALIGN - 1);
830 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
831 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
832 sgen_pin_stage_ptr ((void*)addr);
833 binary_protocol_pin_stage (start, (void*)addr);
834 sgen_pin_stats_register_address ((char*)addr, pin_type);
840 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
844 * The first thing we do in a collection is to identify pinned objects.
845 * This function considers all the areas of memory that need to be
846 * conservatively scanned.
849 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
853 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);
854 /* objects pinned from the API are inside these roots */
855 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
856 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
857 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
858 } SGEN_HASH_TABLE_FOREACH_END;
859 /* now deal with the thread stacks
860 * in the future we should be able to conservatively scan only:
861 * *) the cpu registers
862 * *) the unmanaged stack frames
863 * *) the _last_ managed stack frame
864 * *) pointers slots in managed frames
866 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
870 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
872 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
873 ctx->ops->copy_or_mark_object (obj, ctx->queue);
877 * The memory area from start_root to end_root contains pointers to objects.
878 * Their position is precisely described by @desc (this means that the pointer
879 * can be either NULL or the pointer to the start of an object).
880 * This functions copies them to to_space updates them.
882 * This function is not thread-safe!
885 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
887 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
888 SgenGrayQueue *queue = ctx.queue;
890 switch (desc & ROOT_DESC_TYPE_MASK) {
891 case ROOT_DESC_BITMAP:
892 desc >>= ROOT_DESC_TYPE_SHIFT;
894 if ((desc & 1) && *start_root) {
895 copy_func ((GCObject**)start_root, queue);
896 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
902 case ROOT_DESC_COMPLEX: {
903 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
904 gsize bwords = (*bitmap_data) - 1;
905 void **start_run = start_root;
907 while (bwords-- > 0) {
908 gsize bmap = *bitmap_data++;
909 void **objptr = start_run;
911 if ((bmap & 1) && *objptr) {
912 copy_func ((GCObject**)objptr, queue);
913 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
918 start_run += GC_BITS_PER_WORD;
922 case ROOT_DESC_USER: {
923 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
924 marker (start_root, single_arg_user_copy_or_mark, &ctx);
927 case ROOT_DESC_RUN_LEN:
928 g_assert_not_reached ();
930 g_assert_not_reached ();
935 reset_heap_boundaries (void)
937 lowest_heap_address = ~(mword)0;
938 highest_heap_address = 0;
942 sgen_update_heap_boundaries (mword low, mword high)
947 old = lowest_heap_address;
950 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
953 old = highest_heap_address;
956 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
960 * Allocate and setup the data structures needed to be able to allocate objects
961 * in the nursery. The nursery is stored in nursery_section.
966 GCMemSection *section;
973 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
974 /* later we will alloc a larger area for the nursery but only activate
975 * what we need. The rest will be used as expansion if we have too many pinned
976 * objects in the existing nursery.
978 /* FIXME: handle OOM */
979 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
981 alloc_size = sgen_nursery_size;
983 /* If there isn't enough space even for the nursery we should simply abort. */
984 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
986 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
987 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
988 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 ());
989 section->data = section->next_data = data;
990 section->size = alloc_size;
991 section->end_data = data + sgen_nursery_size;
992 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
993 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
994 section->num_scan_start = scan_starts;
996 nursery_section = section;
998 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
1002 mono_gc_get_logfile (void)
1004 return gc_debug_file;
1008 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1010 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1011 SgenGrayQueue *queue = ctx.queue;
1014 for (i = 0; i < fin_queue->next_slot; ++i) {
1015 GCObject *obj = (GCObject *)fin_queue->data [i];
1018 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1019 copy_func ((GCObject**)&fin_queue->data [i], queue);
1024 generation_name (int generation)
1026 switch (generation) {
1027 case GENERATION_NURSERY: return "nursery";
1028 case GENERATION_OLD: return "old";
1029 default: g_assert_not_reached ();
1034 sgen_generation_name (int generation)
1036 return generation_name (generation);
1040 finish_gray_stack (int generation, ScanCopyContext ctx)
1044 int done_with_ephemerons, ephemeron_rounds = 0;
1045 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1046 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1047 SgenGrayQueue *queue = ctx.queue;
1049 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1051 * We copied all the reachable objects. Now it's the time to copy
1052 * the objects that were not referenced by the roots, but by the copied objects.
1053 * we built a stack of objects pointed to by gray_start: they are
1054 * additional roots and we may add more items as we go.
1055 * We loop until gray_start == gray_objects which means no more objects have
1056 * been added. Note this is iterative: no recursion is involved.
1057 * We need to walk the LO list as well in search of marked big objects
1058 * (use a flag since this is needed only on major collections). We need to loop
1059 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1060 * To achieve better cache locality and cache usage, we drain the gray stack
1061 * frequently, after each object is copied, and just finish the work here.
1063 sgen_drain_gray_stack (ctx);
1065 SGEN_LOG (2, "%s generation done", generation_name (generation));
1068 Reset bridge data, we might have lingering data from a previous collection if this is a major
1069 collection trigged by minor overflow.
1071 We must reset the gathered bridges since their original block might be evacuated due to major
1072 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1074 if (sgen_client_bridge_need_processing ())
1075 sgen_client_bridge_reset_data ();
1078 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1079 * to ensure they see the full set of live objects.
1081 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1084 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1085 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1086 * objects that are in fact reachable.
1088 done_with_ephemerons = 0;
1090 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1091 sgen_drain_gray_stack (ctx);
1093 } while (!done_with_ephemerons);
1095 if (sgen_client_bridge_need_processing ()) {
1096 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1097 sgen_drain_gray_stack (ctx);
1098 sgen_collect_bridge_objects (generation, ctx);
1099 if (generation == GENERATION_OLD)
1100 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1103 Do the first bridge step here, as the collector liveness state will become useless after that.
1105 An important optimization is to only proccess the possibly dead part of the object graph and skip
1106 over all live objects as we transitively know everything they point must be alive too.
1108 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1110 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1111 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1114 sgen_client_bridge_processing_stw_step ();
1118 Make sure we drain the gray stack before processing disappearing links and finalizers.
1119 If we don't make sure it is empty we might wrongly see a live object as dead.
1121 sgen_drain_gray_stack (ctx);
1124 We must clear weak links that don't track resurrection before processing object ready for
1125 finalization so they can be cleared before that.
1127 sgen_null_link_in_range (generation, ctx, FALSE);
1128 if (generation == GENERATION_OLD)
1129 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1132 /* walk the finalization queue and move also the objects that need to be
1133 * finalized: use the finalized objects as new roots so the objects they depend
1134 * on are also not reclaimed. As with the roots above, only objects in the nursery
1135 * are marked/copied.
1137 sgen_finalize_in_range (generation, ctx);
1138 if (generation == GENERATION_OLD)
1139 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1140 /* drain the new stack that might have been created */
1141 SGEN_LOG (6, "Precise scan of gray area post fin");
1142 sgen_drain_gray_stack (ctx);
1145 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1147 done_with_ephemerons = 0;
1149 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1150 sgen_drain_gray_stack (ctx);
1152 } while (!done_with_ephemerons);
1154 sgen_client_clear_unreachable_ephemerons (ctx);
1157 * We clear togglerefs only after all possible chances of revival are done.
1158 * This is semantically more inline with what users expect and it allows for
1159 * user finalizers to correctly interact with TR objects.
1161 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1164 SGEN_LOG (2, "Finalize queue handling scan for %s generation: %lld usecs %d ephemeron rounds", generation_name (generation), TV_ELAPSED (atv, btv), ephemeron_rounds);
1167 * handle disappearing links
1168 * Note we do this after checking the finalization queue because if an object
1169 * survives (at least long enough to be finalized) we don't clear the link.
1170 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1171 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1174 g_assert (sgen_gray_object_queue_is_empty (queue));
1176 sgen_null_link_in_range (generation, ctx, TRUE);
1177 if (generation == GENERATION_OLD)
1178 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1179 if (sgen_gray_object_queue_is_empty (queue))
1181 sgen_drain_gray_stack (ctx);
1184 g_assert (sgen_gray_object_queue_is_empty (queue));
1186 sgen_gray_object_queue_trim_free_list (queue);
1187 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1191 sgen_check_section_scan_starts (GCMemSection *section)
1194 for (i = 0; i < section->num_scan_start; ++i) {
1195 if (section->scan_starts [i]) {
1196 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1197 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1203 check_scan_starts (void)
1205 if (!do_scan_starts_check)
1207 sgen_check_section_scan_starts (nursery_section);
1208 major_collector.check_scan_starts ();
1212 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1216 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1217 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1218 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1219 } SGEN_HASH_TABLE_FOREACH_END;
1225 static gboolean inited = FALSE;
1230 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1232 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1233 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1234 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1235 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1236 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1237 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1239 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1240 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1241 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1242 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1243 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1244 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1245 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1246 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1247 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1248 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1250 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1252 #ifdef HEAVY_STATISTICS
1253 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1254 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1255 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1256 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1257 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1259 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1260 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1262 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1263 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1264 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1265 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1267 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1268 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1270 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1272 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1273 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1274 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1275 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1277 sgen_nursery_allocator_init_heavy_stats ();
1285 reset_pinned_from_failed_allocation (void)
1287 bytes_pinned_from_failed_allocation = 0;
1291 sgen_set_pinned_from_failed_allocation (mword objsize)
1293 bytes_pinned_from_failed_allocation += objsize;
1297 sgen_collection_is_concurrent (void)
1299 switch (current_collection_generation) {
1300 case GENERATION_NURSERY:
1302 case GENERATION_OLD:
1303 return concurrent_collection_in_progress;
1305 g_error ("Invalid current generation %d", current_collection_generation);
1311 sgen_concurrent_collection_in_progress (void)
1313 return concurrent_collection_in_progress;
1317 SgenThreadPoolJob job;
1318 SgenObjectOperations *ops;
1322 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1324 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1325 ScanJob *job_data = (ScanJob*)job;
1326 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1327 remset.scan_remsets (ctx);
1331 SgenThreadPoolJob job;
1332 SgenObjectOperations *ops;
1336 } ScanFromRegisteredRootsJob;
1339 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1341 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1342 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1343 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1345 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1349 SgenThreadPoolJob job;
1350 SgenObjectOperations *ops;
1353 } ScanThreadDataJob;
1356 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1358 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1359 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1360 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1362 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1366 SgenThreadPoolJob job;
1367 SgenObjectOperations *ops;
1368 SgenPointerQueue *queue;
1369 } ScanFinalizerEntriesJob;
1372 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1374 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1375 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1376 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1378 scan_finalizer_entries (job_data->queue, ctx);
1382 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1384 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1385 ScanJob *job_data = (ScanJob*)job;
1386 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1388 g_assert (concurrent_collection_in_progress);
1389 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1393 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1395 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1396 ScanJob *job_data = (ScanJob*)job;
1397 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1399 g_assert (concurrent_collection_in_progress);
1400 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1404 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1406 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1407 ScanJob *job_data = (ScanJob*)job;
1408 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1410 g_assert (concurrent_collection_in_progress);
1412 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1413 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1417 init_gray_queue (gboolean use_workers)
1420 sgen_workers_init_distribute_gray_queue ();
1421 sgen_gray_object_queue_init (&gray_queue, NULL);
1425 enqueue_scan_from_roots_jobs (char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1427 ScanFromRegisteredRootsJob *scrrj;
1428 ScanThreadDataJob *stdj;
1429 ScanFinalizerEntriesJob *sfej;
1431 /* registered roots, this includes static fields */
1433 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1435 scrrj->heap_start = heap_start;
1436 scrrj->heap_end = heap_end;
1437 scrrj->root_type = ROOT_TYPE_NORMAL;
1438 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1440 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1442 scrrj->heap_start = heap_start;
1443 scrrj->heap_end = heap_end;
1444 scrrj->root_type = ROOT_TYPE_WBARRIER;
1445 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1449 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1450 stdj->heap_start = heap_start;
1451 stdj->heap_end = heap_end;
1452 sgen_workers_enqueue_job (&stdj->job, enqueue);
1454 /* Scan the list of objects ready for finalization. */
1456 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1457 sfej->queue = &fin_ready_queue;
1459 sgen_workers_enqueue_job (&sfej->job, enqueue);
1461 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1462 sfej->queue = &critical_fin_queue;
1464 sgen_workers_enqueue_job (&sfej->job, enqueue);
1468 * Perform a nursery collection.
1470 * Return whether any objects were late-pinned due to being out of memory.
1473 collect_nursery (SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1475 gboolean needs_major;
1476 size_t max_garbage_amount;
1478 mword fragment_total;
1480 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1481 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue);
1485 if (disable_minor_collections)
1488 TV_GETTIME (last_minor_collection_start_tv);
1489 atv = last_minor_collection_start_tv;
1491 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1493 if (do_verify_nursery || do_dump_nursery_content)
1494 sgen_debug_verify_nursery (do_dump_nursery_content);
1496 current_collection_generation = GENERATION_NURSERY;
1498 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1500 reset_pinned_from_failed_allocation ();
1502 check_scan_starts ();
1504 sgen_nursery_alloc_prepare_for_minor ();
1508 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1509 /* FIXME: optimize later to use the higher address where an object can be present */
1510 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1512 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 ()));
1513 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1514 g_assert (nursery_section->size >= max_garbage_amount);
1516 /* world must be stopped already */
1518 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1520 sgen_client_pre_collection_checks ();
1522 nursery_section->next_data = nursery_next;
1524 major_collector.start_nursery_collection ();
1526 sgen_memgov_minor_collection_start ();
1528 init_gray_queue (FALSE);
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);
1536 if (consistency_check_at_minor_collection)
1537 sgen_check_consistency ();
1539 sgen_process_fin_stage_entries ();
1541 /* pin from pinned handles */
1542 sgen_init_pinning ();
1543 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1544 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1545 /* pin cemented objects */
1546 sgen_pin_cemented_objects ();
1547 /* identify pinned objects */
1548 sgen_optimize_pin_queue ();
1549 sgen_pinning_setup_section (nursery_section);
1551 pin_objects_in_nursery (FALSE, ctx);
1552 sgen_pinning_trim_queue_to_section (nursery_section);
1555 time_minor_pinning += TV_ELAPSED (btv, atv);
1556 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), TV_ELAPSED (btv, atv));
1557 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1559 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1560 sj->ops = object_ops;
1561 sgen_workers_enqueue_job (&sj->job, FALSE);
1563 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1565 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1566 SGEN_LOG (2, "Old generation scan: %lld usecs", TV_ELAPSED (atv, btv));
1568 sgen_pin_stats_print_class_stats ();
1570 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1571 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1574 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1576 enqueue_scan_from_roots_jobs (sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1579 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1581 finish_gray_stack (GENERATION_NURSERY, ctx);
1584 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1585 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1587 if (objects_pinned) {
1588 sgen_optimize_pin_queue ();
1589 sgen_pinning_setup_section (nursery_section);
1592 /* walk the pin_queue, build up the fragment list of free memory, unmark
1593 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1596 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1597 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1598 if (!fragment_total)
1601 /* Clear TLABs for all threads */
1602 sgen_clear_tlabs ();
1604 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1606 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1607 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1609 if (consistency_check_at_minor_collection)
1610 sgen_check_major_refs ();
1612 major_collector.finish_nursery_collection ();
1614 TV_GETTIME (last_minor_collection_end_tv);
1615 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1617 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1619 /* prepare the pin queue for the next collection */
1620 sgen_finish_pinning ();
1621 if (sgen_have_pending_finalizers ()) {
1622 SGEN_LOG (4, "Finalizer-thread wakeup");
1623 sgen_client_finalize_notify ();
1625 sgen_pin_stats_reset ();
1626 /* clear cemented hash */
1627 sgen_cement_clear_below_threshold ();
1629 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
1631 remset.finish_minor_collection ();
1633 check_scan_starts ();
1635 binary_protocol_flush_buffers (FALSE);
1637 sgen_memgov_minor_collection_end ();
1639 /*objects are late pinned because of lack of memory, so a major is a good call*/
1640 needs_major = objects_pinned > 0;
1641 current_collection_generation = -1;
1644 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1646 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1647 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1653 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1654 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1655 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1656 } CopyOrMarkFromRootsMode;
1659 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1664 /* FIXME: only use these values for the precise scan
1665 * note that to_space pointers should be excluded anyway...
1667 char *heap_start = NULL;
1668 char *heap_end = (char*)-1;
1669 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, WORKERS_DISTRIBUTE_GRAY_QUEUE);
1670 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1672 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1674 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1675 /*This cleans up unused fragments */
1676 sgen_nursery_allocator_prepare_for_pinning ();
1678 if (do_concurrent_checks)
1679 sgen_debug_check_nursery_is_clean ();
1681 /* The concurrent collector doesn't touch the nursery. */
1682 sgen_nursery_alloc_prepare_for_major ();
1685 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1689 /* Pinning depends on this */
1690 sgen_clear_nursery_fragments ();
1692 if (whole_heap_check_before_collection)
1693 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1696 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1698 if (!sgen_collection_is_concurrent ())
1699 nursery_section->next_data = sgen_get_nursery_end ();
1700 /* we should also coalesce scanning from sections close to each other
1701 * and deal with pointers outside of the sections later.
1706 sgen_client_pre_collection_checks ();
1708 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1709 /* Remsets are not useful for a major collection */
1710 remset.clear_cards ();
1713 sgen_process_fin_stage_entries ();
1716 sgen_init_pinning ();
1717 SGEN_LOG (6, "Collecting pinned addresses");
1718 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1719 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1720 /* Pin cemented objects that were forced */
1721 sgen_pin_cemented_objects ();
1723 sgen_optimize_pin_queue ();
1724 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1726 * Cemented objects that are in the pinned list will be marked. When
1727 * marking concurrently we won't mark mod-union cards for these objects.
1728 * Instead they will remain cemented until the next major collection,
1729 * when we will recheck if they are still pinned in the roots.
1731 sgen_cement_force_pinned ();
1734 sgen_client_collecting_major_1 ();
1737 * pin_queue now contains all candidate pointers, sorted and
1738 * uniqued. We must do two passes now to figure out which
1739 * objects are pinned.
1741 * The first is to find within the pin_queue the area for each
1742 * section. This requires that the pin_queue be sorted. We
1743 * also process the LOS objects and pinned chunks here.
1745 * The second, destructive, pass is to reduce the section
1746 * areas to pointers to the actually pinned objects.
1748 SGEN_LOG (6, "Pinning from sections");
1749 /* first pass for the sections */
1750 sgen_find_section_pin_queue_start_end (nursery_section);
1751 /* identify possible pointers to the insize of large objects */
1752 SGEN_LOG (6, "Pinning from large objects");
1753 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1755 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1756 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1758 if (sgen_los_object_is_pinned (bigobj->data)) {
1759 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1762 sgen_los_pin_object (bigobj->data);
1763 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1764 GRAY_OBJECT_ENQUEUE (WORKERS_DISTRIBUTE_GRAY_QUEUE, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1765 sgen_pin_stats_register_object (bigobj->data, safe_object_get_size (bigobj->data));
1766 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1767 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1768 (unsigned long)sgen_los_object_size (bigobj));
1770 sgen_client_pinned_los_object (bigobj->data);
1774 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1775 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1776 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1778 major_collector.pin_objects (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1779 if (old_next_pin_slot)
1780 *old_next_pin_slot = sgen_get_pinned_count ();
1783 time_major_pinning += TV_ELAPSED (atv, btv);
1784 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), TV_ELAPSED (atv, btv));
1785 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1787 major_collector.init_to_space ();
1789 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1790 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1791 if (sgen_workers_have_idle_work ()) {
1793 * We force the finish of the worker with the new object ops context
1794 * which can also do copying. We need to have finished pinning.
1796 sgen_workers_start_all_workers (object_ops, NULL);
1797 sgen_workers_join ();
1801 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1802 main_gc_thread = mono_native_thread_self ();
1805 sgen_client_collecting_major_2 ();
1808 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1810 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1812 enqueue_scan_from_roots_jobs (heap_start, heap_end, object_ops, FALSE);
1815 time_major_scan_roots += TV_ELAPSED (atv, btv);
1818 * We start the concurrent worker after pinning and after we scanned the roots
1819 * in order to make sure that the worker does not finish before handling all
1822 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1823 if (precleaning_enabled) {
1825 /* Mod union preclean job */
1826 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1827 sj->ops = object_ops;
1828 sgen_workers_start_all_workers (object_ops, &sj->job);
1830 sgen_workers_start_all_workers (object_ops, NULL);
1832 gray_queue_enable_redirect (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1835 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1838 /* Mod union card table */
1839 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1840 sj->ops = object_ops;
1841 sgen_workers_enqueue_job (&sj->job, FALSE);
1843 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1844 sj->ops = object_ops;
1845 sgen_workers_enqueue_job (&sj->job, FALSE);
1848 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1851 sgen_pin_stats_print_class_stats ();
1855 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1857 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1858 sgen_finish_pinning ();
1860 sgen_pin_stats_reset ();
1862 if (do_concurrent_checks)
1863 sgen_debug_check_nursery_is_clean ();
1868 major_start_collection (gboolean concurrent, size_t *old_next_pin_slot)
1870 SgenObjectOperations *object_ops;
1872 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1874 current_collection_generation = GENERATION_OLD;
1876 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1879 sgen_cement_reset ();
1882 g_assert (major_collector.is_concurrent);
1883 concurrent_collection_in_progress = TRUE;
1885 object_ops = &major_collector.major_ops_concurrent_start;
1887 object_ops = &major_collector.major_ops_serial;
1890 reset_pinned_from_failed_allocation ();
1892 sgen_memgov_major_collection_start ();
1894 //count_ref_nonref_objs ();
1895 //consistency_check ();
1897 check_scan_starts ();
1900 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1901 gc_stats.major_gc_count ++;
1903 if (major_collector.start_major_collection)
1904 major_collector.start_major_collection ();
1906 major_copy_or_mark_from_roots (old_next_pin_slot, concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL, object_ops);
1907 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1911 major_finish_collection (const char *reason, size_t old_next_pin_slot, gboolean forced)
1913 ScannedObjectCounts counts;
1914 SgenObjectOperations *object_ops;
1915 mword fragment_total;
1921 if (concurrent_collection_in_progress) {
1922 object_ops = &major_collector.major_ops_concurrent_finish;
1924 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1926 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1928 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1929 main_gc_thread = NULL;
1932 object_ops = &major_collector.major_ops_serial;
1935 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1937 /* all the objects in the heap */
1938 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue));
1940 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1942 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1944 if (objects_pinned) {
1945 g_assert (!concurrent_collection_in_progress);
1948 * This is slow, but we just OOM'd.
1950 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1951 * queue is laid out at this point.
1953 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1955 * We need to reestablish all pinned nursery objects in the pin queue
1956 * because they're needed for fragment creation. Unpinning happens by
1957 * walking the whole queue, so it's not necessary to reestablish where major
1958 * heap block pins are - all we care is that they're still in there
1961 sgen_optimize_pin_queue ();
1962 sgen_find_section_pin_queue_start_end (nursery_section);
1966 reset_heap_boundaries ();
1967 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1969 /* walk the pin_queue, build up the fragment list of free memory, unmark
1970 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1973 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1974 if (!fragment_total)
1976 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1978 if (do_concurrent_checks && concurrent_collection_in_progress)
1979 sgen_debug_check_nursery_is_clean ();
1981 /* prepare the pin queue for the next collection */
1982 sgen_finish_pinning ();
1984 /* Clear TLABs for all threads */
1985 sgen_clear_tlabs ();
1987 sgen_pin_stats_reset ();
1989 sgen_cement_clear_below_threshold ();
1991 if (check_mark_bits_after_major_collection)
1992 sgen_check_heap_marked (concurrent_collection_in_progress);
1995 time_major_fragment_creation += TV_ELAPSED (atv, btv);
1997 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
1998 sgen_memgov_major_pre_sweep ();
2001 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2006 time_major_los_sweep += TV_ELAPSED (atv, btv);
2008 major_collector.sweep ();
2010 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2013 time_major_sweep += TV_ELAPSED (btv, atv);
2015 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2017 if (sgen_have_pending_finalizers ()) {
2018 SGEN_LOG (4, "Finalizer-thread wakeup");
2019 sgen_client_finalize_notify ();
2022 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2024 sgen_memgov_major_collection_end (forced);
2025 current_collection_generation = -1;
2027 memset (&counts, 0, sizeof (ScannedObjectCounts));
2028 major_collector.finish_major_collection (&counts);
2030 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
2032 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2033 if (concurrent_collection_in_progress)
2034 concurrent_collection_in_progress = FALSE;
2036 check_scan_starts ();
2038 binary_protocol_flush_buffers (FALSE);
2040 //consistency_check ();
2042 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2046 major_do_collection (const char *reason, gboolean forced)
2048 TV_DECLARE (time_start);
2049 TV_DECLARE (time_end);
2050 size_t old_next_pin_slot;
2052 if (disable_major_collections)
2055 if (major_collector.get_and_reset_num_major_objects_marked) {
2056 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2057 g_assert (!num_marked);
2060 /* world must be stopped already */
2061 TV_GETTIME (time_start);
2063 major_start_collection (FALSE, &old_next_pin_slot);
2064 major_finish_collection (reason, old_next_pin_slot, forced);
2066 TV_GETTIME (time_end);
2067 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2069 /* FIXME: also report this to the user, preferably in gc-end. */
2070 if (major_collector.get_and_reset_num_major_objects_marked)
2071 major_collector.get_and_reset_num_major_objects_marked ();
2073 return bytes_pinned_from_failed_allocation > 0;
2077 major_start_concurrent_collection (const char *reason)
2079 TV_DECLARE (time_start);
2080 TV_DECLARE (time_end);
2081 long long num_objects_marked;
2083 if (disable_major_collections)
2086 TV_GETTIME (time_start);
2087 SGEN_TV_GETTIME (time_major_conc_collection_start);
2089 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2090 g_assert (num_objects_marked == 0);
2092 binary_protocol_concurrent_start ();
2094 // FIXME: store reason and pass it when finishing
2095 major_start_collection (TRUE, NULL);
2097 gray_queue_redirect (&gray_queue);
2099 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2101 TV_GETTIME (time_end);
2102 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2104 current_collection_generation = -1;
2108 * Returns whether the major collection has finished.
2111 major_should_finish_concurrent_collection (void)
2113 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2114 return sgen_workers_all_done ();
2118 major_update_concurrent_collection (void)
2120 TV_DECLARE (total_start);
2121 TV_DECLARE (total_end);
2123 TV_GETTIME (total_start);
2125 binary_protocol_concurrent_update ();
2127 major_collector.update_cardtable_mod_union ();
2128 sgen_los_update_cardtable_mod_union ();
2130 TV_GETTIME (total_end);
2131 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2135 major_finish_concurrent_collection (gboolean forced)
2137 TV_DECLARE (total_start);
2138 TV_DECLARE (total_end);
2140 TV_GETTIME (total_start);
2142 binary_protocol_concurrent_finish ();
2145 * We need to stop all workers since we're updating the cardtable below.
2146 * The workers will be resumed with a finishing pause context to avoid
2147 * additional cardtable and object scanning.
2149 sgen_workers_stop_all_workers ();
2151 SGEN_TV_GETTIME (time_major_conc_collection_end);
2152 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2154 major_collector.update_cardtable_mod_union ();
2155 sgen_los_update_cardtable_mod_union ();
2157 if (mod_union_consistency_check)
2158 sgen_check_mod_union_consistency ();
2160 current_collection_generation = GENERATION_OLD;
2161 sgen_cement_reset ();
2162 major_finish_collection ("finishing", -1, forced);
2164 if (whole_heap_check_before_collection)
2165 sgen_check_whole_heap (FALSE);
2167 TV_GETTIME (total_end);
2168 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2170 current_collection_generation = -1;
2174 * Ensure an allocation request for @size will succeed by freeing enough memory.
2176 * LOCKING: The GC lock MUST be held.
2179 sgen_ensure_free_space (size_t size, int generation)
2181 int generation_to_collect = -1;
2182 const char *reason = NULL;
2184 if (generation == GENERATION_OLD) {
2185 if (sgen_need_major_collection (size)) {
2186 reason = "LOS overflow";
2187 generation_to_collect = GENERATION_OLD;
2190 if (degraded_mode) {
2191 if (sgen_need_major_collection (size)) {
2192 reason = "Degraded mode overflow";
2193 generation_to_collect = GENERATION_OLD;
2195 } else if (sgen_need_major_collection (size)) {
2196 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2197 generation_to_collect = GENERATION_OLD;
2199 generation_to_collect = GENERATION_NURSERY;
2200 reason = "Nursery full";
2204 if (generation_to_collect == -1) {
2205 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2206 generation_to_collect = GENERATION_OLD;
2207 reason = "Finish concurrent collection";
2211 if (generation_to_collect == -1)
2213 sgen_perform_collection (size, generation_to_collect, reason, FALSE);
2217 * LOCKING: Assumes the GC lock is held.
2220 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish)
2222 TV_DECLARE (gc_start);
2223 TV_DECLARE (gc_end);
2224 TV_DECLARE (gc_total_start);
2225 TV_DECLARE (gc_total_end);
2226 GGTimingInfo infos [2];
2227 int overflow_generation_to_collect = -1;
2228 int oldest_generation_collected = generation_to_collect;
2229 const char *overflow_reason = NULL;
2231 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2233 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2235 TV_GETTIME (gc_start);
2237 sgen_stop_world (generation_to_collect);
2239 TV_GETTIME (gc_total_start);
2241 if (concurrent_collection_in_progress) {
2243 * If the concurrent worker is finished or we are asked to do a major collection
2244 * then we finish the concurrent collection.
2246 gboolean finish = major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD;
2249 major_finish_concurrent_collection (wait_to_finish);
2250 oldest_generation_collected = GENERATION_OLD;
2252 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY, "Why aren't we finishing the concurrent collection?");
2253 major_update_concurrent_collection ();
2254 collect_nursery (NULL, FALSE);
2260 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2263 * There's no concurrent collection in progress. Collect the generation we're asked
2264 * to collect. If the major collector is concurrent and we're not forced to wait,
2265 * start a concurrent collection.
2267 // FIXME: extract overflow reason
2268 if (generation_to_collect == GENERATION_NURSERY) {
2269 if (collect_nursery (NULL, FALSE)) {
2270 overflow_generation_to_collect = GENERATION_OLD;
2271 overflow_reason = "Minor overflow";
2274 if (major_collector.is_concurrent && !wait_to_finish) {
2275 collect_nursery (NULL, FALSE);
2276 major_start_concurrent_collection (reason);
2277 // FIXME: set infos[0] properly
2281 if (major_do_collection (reason, wait_to_finish)) {
2282 overflow_generation_to_collect = GENERATION_NURSERY;
2283 overflow_reason = "Excessive pinning";
2287 TV_GETTIME (gc_end);
2289 memset (infos, 0, sizeof (infos));
2290 infos [0].generation = generation_to_collect;
2291 infos [0].reason = reason;
2292 infos [0].is_overflow = FALSE;
2293 infos [1].generation = -1;
2294 infos [0].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2296 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2298 if (overflow_generation_to_collect != -1) {
2300 * We need to do an overflow collection, either because we ran out of memory
2301 * or the nursery is fully pinned.
2304 infos [1].generation = overflow_generation_to_collect;
2305 infos [1].reason = overflow_reason;
2306 infos [1].is_overflow = TRUE;
2309 if (overflow_generation_to_collect == GENERATION_NURSERY)
2310 collect_nursery (NULL, FALSE);
2312 major_do_collection (overflow_reason, wait_to_finish);
2314 TV_GETTIME (gc_end);
2315 infos [1].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2317 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2320 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2322 /* this also sets the proper pointers for the next allocation */
2323 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2324 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2325 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2326 sgen_dump_pin_queue ();
2331 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2333 TV_GETTIME (gc_total_end);
2334 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2336 sgen_restart_world (oldest_generation_collected, infos);
2340 * ######################################################################
2341 * ######## Memory allocation from the OS
2342 * ######################################################################
2343 * This section of code deals with getting memory from the OS and
2344 * allocating memory for GC-internal data structures.
2345 * Internal memory can be handled with a freelist for small objects.
2351 G_GNUC_UNUSED static void
2352 report_internal_mem_usage (void)
2354 printf ("Internal memory usage:\n");
2355 sgen_report_internal_mem_usage ();
2356 printf ("Pinned memory usage:\n");
2357 major_collector.report_pinned_memory_usage ();
2361 * ######################################################################
2362 * ######## Finalization support
2363 * ######################################################################
2367 * If the object has been forwarded it means it's still referenced from a root.
2368 * If it is pinned it's still alive as well.
2369 * A LOS object is only alive if we have pinned it.
2370 * Return TRUE if @obj is ready to be finalized.
2372 static inline gboolean
2373 sgen_is_object_alive (GCObject *object)
2375 if (ptr_in_nursery (object))
2376 return sgen_nursery_is_object_alive (object);
2378 return sgen_major_is_object_alive (object);
2382 * This function returns true if @object is either alive and belongs to the
2383 * current collection - major collections are full heap, so old gen objects
2384 * are never alive during a minor collection.
2387 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2389 if (ptr_in_nursery (object))
2390 return sgen_nursery_is_object_alive (object);
2392 if (current_collection_generation == GENERATION_NURSERY)
2395 return sgen_major_is_object_alive (object);
2400 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2402 return !sgen_is_object_alive (object);
2406 sgen_queue_finalization_entry (GCObject *obj)
2408 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2410 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2412 sgen_client_object_queued_for_finalization (obj);
2416 sgen_object_is_live (GCObject *obj)
2418 return sgen_is_object_alive_and_on_current_collection (obj);
2422 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2423 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2424 * all finalizers have really finished running.
2426 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2427 * This means that just checking whether the queues are empty leaves the possibility that an
2428 * object might have been dequeued but not yet finalized. That's why we need the additional
2429 * flag `pending_unqueued_finalizer`.
2432 static volatile gboolean pending_unqueued_finalizer = FALSE;
2435 sgen_gc_invoke_finalizers (void)
2439 g_assert (!pending_unqueued_finalizer);
2441 /* FIXME: batch to reduce lock contention */
2442 while (sgen_have_pending_finalizers ()) {
2448 * We need to set `pending_unqueued_finalizer` before dequeing the
2449 * finalizable object.
2451 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2452 pending_unqueued_finalizer = TRUE;
2453 mono_memory_write_barrier ();
2454 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2455 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2456 pending_unqueued_finalizer = TRUE;
2457 mono_memory_write_barrier ();
2458 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2464 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2472 /* the object is on the stack so it is pinned */
2473 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2474 sgen_client_run_finalize (obj);
2477 if (pending_unqueued_finalizer) {
2478 mono_memory_write_barrier ();
2479 pending_unqueued_finalizer = FALSE;
2486 sgen_have_pending_finalizers (void)
2488 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2492 * ######################################################################
2493 * ######## registered roots support
2494 * ######################################################################
2498 * We do not coalesce roots.
2501 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2503 RootRecord new_root;
2506 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2507 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2508 /* we allow changing the size and the descriptor (for thread statics etc) */
2510 size_t old_size = root->end_root - start;
2511 root->end_root = start + size;
2512 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2513 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2514 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2515 root->root_desc = descr;
2517 roots_size -= old_size;
2523 new_root.end_root = start + size;
2524 new_root.root_desc = descr;
2525 new_root.source = source;
2528 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2531 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);
2538 sgen_deregister_root (char* addr)
2544 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2545 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2546 roots_size -= (root.end_root - addr);
2552 * ######################################################################
2553 * ######## Thread handling (stop/start code)
2554 * ######################################################################
2558 sgen_get_current_collection_generation (void)
2560 return current_collection_generation;
2564 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2566 #ifndef HAVE_KW_THREAD
2567 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2570 sgen_init_tlab_info (info);
2572 sgen_client_thread_register (info, stack_bottom_fallback);
2578 sgen_thread_unregister (SgenThreadInfo *p)
2580 sgen_client_thread_unregister (p);
2584 * ######################################################################
2585 * ######## Write barriers
2586 * ######################################################################
2590 * Note: the write barriers first do the needed GC work and then do the actual store:
2591 * this way the value is visible to the conservative GC scan after the write barrier
2592 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2593 * the conservative scan, otherwise by the remembered set scan.
2597 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2599 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2600 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2601 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2602 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2606 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2607 if (binary_protocol_is_heavy_enabled ()) {
2609 for (i = 0; i < count; ++i) {
2610 gpointer dest = (gpointer*)dest_ptr + i;
2611 gpointer obj = *((gpointer*)src_ptr + i);
2613 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2618 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2622 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2626 HEAVY_STAT (++stat_wbarrier_generic_store);
2628 sgen_client_wbarrier_generic_nostore_check (ptr);
2630 obj = *(gpointer*)ptr;
2632 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2635 * We need to record old->old pointer locations for the
2636 * concurrent collector.
2638 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2639 SGEN_LOG (8, "Skipping remset at %p", ptr);
2643 SGEN_LOG (8, "Adding remset at %p", ptr);
2645 remset.wbarrier_generic_nostore (ptr);
2649 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2651 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2652 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2653 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2654 mono_gc_wbarrier_generic_nostore (ptr);
2655 sgen_dummy_use (value);
2658 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2659 * as an atomic operation with release semantics.
2662 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2664 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2666 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2668 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2670 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2671 mono_gc_wbarrier_generic_nostore (ptr);
2673 sgen_dummy_use (value);
2677 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2679 GCObject **dest = (GCObject **)_dest;
2680 GCObject **src = (GCObject **)_src;
2684 mono_gc_wbarrier_generic_store (dest, *src);
2689 size -= SIZEOF_VOID_P;
2695 * ######################################################################
2696 * ######## Other mono public interface functions.
2697 * ######################################################################
2701 sgen_gc_collect (int generation)
2706 sgen_perform_collection (0, generation, "user request", TRUE);
2711 sgen_gc_collection_count (int generation)
2713 if (generation == 0)
2714 return gc_stats.minor_gc_count;
2715 return gc_stats.major_gc_count;
2719 sgen_gc_get_used_size (void)
2723 tot = los_memory_usage;
2724 tot += nursery_section->next_data - nursery_section->data;
2725 tot += major_collector.get_used_size ();
2726 /* FIXME: account for pinned objects */
2732 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2736 va_start (ap, description_format);
2738 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2739 vfprintf (stderr, description_format, ap);
2741 fprintf (stderr, " - %s", fallback);
2742 fprintf (stderr, "\n");
2748 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2751 double val = strtod (opt, &endptr);
2752 if (endptr == opt) {
2753 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2756 else if (val < min || val > max) {
2757 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2769 char *major_collector_opt = NULL;
2770 char *minor_collector_opt = NULL;
2771 size_t max_heap = 0;
2772 size_t soft_limit = 0;
2774 gboolean debug_print_allowance = FALSE;
2775 double allowance_ratio = 0, save_target = 0;
2776 gboolean cement_enabled = TRUE;
2779 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2782 /* already inited */
2785 /* being inited by another thread */
2786 mono_thread_info_usleep (1000);
2789 /* we will init it */
2792 g_assert_not_reached ();
2794 } while (result != 0);
2796 SGEN_TV_GETTIME (sgen_init_timestamp);
2798 #ifdef SGEN_WITHOUT_MONO
2799 mono_thread_smr_init ();
2802 mono_coop_mutex_init (&gc_mutex);
2804 gc_debug_file = stderr;
2806 mono_coop_mutex_init (&sgen_interruption_mutex);
2808 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2809 opts = g_strsplit (env, ",", -1);
2810 for (ptr = opts; *ptr; ++ptr) {
2812 if (g_str_has_prefix (opt, "major=")) {
2813 opt = strchr (opt, '=') + 1;
2814 major_collector_opt = g_strdup (opt);
2815 } else if (g_str_has_prefix (opt, "minor=")) {
2816 opt = strchr (opt, '=') + 1;
2817 minor_collector_opt = g_strdup (opt);
2825 sgen_init_internal_allocator ();
2826 sgen_init_nursery_allocator ();
2827 sgen_init_fin_weak_hash ();
2828 sgen_init_hash_table ();
2829 sgen_init_descriptors ();
2830 sgen_init_gray_queues ();
2831 sgen_init_allocator ();
2832 sgen_init_gchandles ();
2834 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2835 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2837 sgen_client_init ();
2839 if (!minor_collector_opt) {
2840 sgen_simple_nursery_init (&sgen_minor_collector);
2842 if (!strcmp (minor_collector_opt, "simple")) {
2844 sgen_simple_nursery_init (&sgen_minor_collector);
2845 } else if (!strcmp (minor_collector_opt, "split")) {
2846 sgen_split_nursery_init (&sgen_minor_collector);
2848 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2849 goto use_simple_nursery;
2853 if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep")) {
2854 use_marksweep_major:
2855 sgen_marksweep_init (&major_collector);
2856 } else if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep-conc")) {
2857 sgen_marksweep_conc_init (&major_collector);
2859 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `marksweep` instead.", "Unknown major collector `%s'.", major_collector_opt);
2860 goto use_marksweep_major;
2863 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2866 gboolean usage_printed = FALSE;
2868 for (ptr = opts; *ptr; ++ptr) {
2870 if (!strcmp (opt, ""))
2872 if (g_str_has_prefix (opt, "major="))
2874 if (g_str_has_prefix (opt, "minor="))
2876 if (g_str_has_prefix (opt, "max-heap-size=")) {
2877 size_t page_size = mono_pagesize ();
2878 size_t max_heap_candidate = 0;
2879 opt = strchr (opt, '=') + 1;
2880 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2881 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2882 if (max_heap != max_heap_candidate)
2883 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2885 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2889 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2890 opt = strchr (opt, '=') + 1;
2891 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2892 if (soft_limit <= 0) {
2893 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2897 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2903 if (g_str_has_prefix (opt, "nursery-size=")) {
2905 opt = strchr (opt, '=') + 1;
2906 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2907 if ((val & (val - 1))) {
2908 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2912 if (val < SGEN_MAX_NURSERY_WASTE) {
2913 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2914 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2918 sgen_nursery_size = val;
2919 sgen_nursery_bits = 0;
2920 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2923 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2929 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2931 opt = strchr (opt, '=') + 1;
2932 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2933 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2938 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2940 opt = strchr (opt, '=') + 1;
2941 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2942 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2943 allowance_ratio = val;
2948 if (!strcmp (opt, "cementing")) {
2949 cement_enabled = TRUE;
2952 if (!strcmp (opt, "no-cementing")) {
2953 cement_enabled = FALSE;
2957 if (!strcmp (opt, "precleaning")) {
2958 precleaning_enabled = TRUE;
2961 if (!strcmp (opt, "no-precleaning")) {
2962 precleaning_enabled = FALSE;
2966 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2969 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2972 if (sgen_client_handle_gc_param (opt))
2975 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2980 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2981 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2982 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2983 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2984 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2985 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2986 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2987 fprintf (stderr, " [no-]cementing\n");
2988 if (major_collector.print_gc_param_usage)
2989 major_collector.print_gc_param_usage ();
2990 if (sgen_minor_collector.print_gc_param_usage)
2991 sgen_minor_collector.print_gc_param_usage ();
2992 sgen_client_print_gc_params_usage ();
2993 fprintf (stderr, " Experimental options:\n");
2994 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2995 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);
2996 fprintf (stderr, "\n");
2998 usage_printed = TRUE;
3003 if (major_collector_opt)
3004 g_free (major_collector_opt);
3006 if (minor_collector_opt)
3007 g_free (minor_collector_opt);
3011 sgen_cement_init (cement_enabled);
3013 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
3014 gboolean usage_printed = FALSE;
3016 opts = g_strsplit (env, ",", -1);
3017 for (ptr = opts; ptr && *ptr; ptr ++) {
3019 if (!strcmp (opt, ""))
3021 if (opt [0] >= '0' && opt [0] <= '9') {
3022 gc_debug_level = atoi (opt);
3027 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3028 gc_debug_file = fopen (rf, "wb");
3030 gc_debug_file = stderr;
3033 } else if (!strcmp (opt, "print-allowance")) {
3034 debug_print_allowance = TRUE;
3035 } else if (!strcmp (opt, "print-pinning")) {
3036 sgen_pin_stats_enable ();
3037 } else if (!strcmp (opt, "verify-before-allocs")) {
3038 verify_before_allocs = 1;
3039 has_per_allocation_action = TRUE;
3040 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3041 char *arg = strchr (opt, '=') + 1;
3042 verify_before_allocs = atoi (arg);
3043 has_per_allocation_action = TRUE;
3044 } else if (!strcmp (opt, "collect-before-allocs")) {
3045 collect_before_allocs = 1;
3046 has_per_allocation_action = TRUE;
3047 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3048 char *arg = strchr (opt, '=') + 1;
3049 has_per_allocation_action = TRUE;
3050 collect_before_allocs = atoi (arg);
3051 } else if (!strcmp (opt, "verify-before-collections")) {
3052 whole_heap_check_before_collection = TRUE;
3053 } else if (!strcmp (opt, "check-at-minor-collections")) {
3054 consistency_check_at_minor_collection = TRUE;
3055 nursery_clear_policy = CLEAR_AT_GC;
3056 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3057 if (!major_collector.is_concurrent) {
3058 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3061 mod_union_consistency_check = TRUE;
3062 } else if (!strcmp (opt, "check-mark-bits")) {
3063 check_mark_bits_after_major_collection = TRUE;
3064 } else if (!strcmp (opt, "check-nursery-pinned")) {
3065 check_nursery_objects_pinned = TRUE;
3066 } else if (!strcmp (opt, "clear-at-gc")) {
3067 nursery_clear_policy = CLEAR_AT_GC;
3068 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3069 nursery_clear_policy = CLEAR_AT_GC;
3070 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3071 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3072 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3073 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3074 } else if (!strcmp (opt, "check-scan-starts")) {
3075 do_scan_starts_check = TRUE;
3076 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3077 do_verify_nursery = TRUE;
3078 } else if (!strcmp (opt, "check-concurrent")) {
3079 if (!major_collector.is_concurrent) {
3080 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3083 nursery_clear_policy = CLEAR_AT_GC;
3084 do_concurrent_checks = TRUE;
3085 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3086 do_dump_nursery_content = TRUE;
3087 } else if (!strcmp (opt, "disable-minor")) {
3088 disable_minor_collections = TRUE;
3089 } else if (!strcmp (opt, "disable-major")) {
3090 disable_major_collections = TRUE;
3091 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3092 char *filename = strchr (opt, '=') + 1;
3093 nursery_clear_policy = CLEAR_AT_GC;
3094 sgen_debug_enable_heap_dump (filename);
3095 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3096 char *filename = strchr (opt, '=') + 1;
3097 char *colon = strrchr (filename, ':');
3100 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3101 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3106 binary_protocol_init (filename, (long long)limit);
3107 } else if (!strcmp (opt, "nursery-canaries")) {
3108 do_verify_nursery = TRUE;
3109 enable_nursery_canaries = TRUE;
3110 } else if (!sgen_client_handle_gc_debug (opt)) {
3111 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3116 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);
3117 fprintf (stderr, "Valid <option>s are:\n");
3118 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3119 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3120 fprintf (stderr, " check-at-minor-collections\n");
3121 fprintf (stderr, " check-mark-bits\n");
3122 fprintf (stderr, " check-nursery-pinned\n");
3123 fprintf (stderr, " verify-before-collections\n");
3124 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3125 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3126 fprintf (stderr, " disable-minor\n");
3127 fprintf (stderr, " disable-major\n");
3128 fprintf (stderr, " check-concurrent\n");
3129 fprintf (stderr, " clear-[nursery-]at-gc\n");
3130 fprintf (stderr, " clear-at-tlab-creation\n");
3131 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3132 fprintf (stderr, " check-scan-starts\n");
3133 fprintf (stderr, " print-allowance\n");
3134 fprintf (stderr, " print-pinning\n");
3135 fprintf (stderr, " heap-dump=<filename>\n");
3136 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3137 fprintf (stderr, " nursery-canaries\n");
3138 sgen_client_print_gc_debug_usage ();
3139 fprintf (stderr, "\n");
3141 usage_printed = TRUE;
3147 if (check_mark_bits_after_major_collection)
3148 nursery_clear_policy = CLEAR_AT_GC;
3150 if (major_collector.post_param_init)
3151 major_collector.post_param_init (&major_collector);
3153 if (major_collector.needs_thread_pool)
3154 sgen_workers_init (1);
3156 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3158 memset (&remset, 0, sizeof (remset));
3160 sgen_card_table_init (&remset);
3162 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");
3168 sgen_get_nursery_clear_policy (void)
3170 return nursery_clear_policy;
3176 mono_coop_mutex_lock (&gc_mutex);
3180 sgen_gc_unlock (void)
3182 mono_coop_mutex_unlock (&gc_mutex);
3186 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3188 major_collector.iterate_live_block_ranges (callback);
3192 sgen_get_major_collector (void)
3194 return &major_collector;
3198 sgen_get_remset (void)
3204 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3206 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3207 sgen_los_count_cards (los_total, los_marked);
3210 static gboolean world_is_stopped = FALSE;
3212 /* LOCKING: assumes the GC lock is held */
3214 sgen_stop_world (int generation)
3216 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3218 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3220 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3222 sgen_client_stop_world (generation);
3224 world_is_stopped = TRUE;
3226 if (binary_protocol_is_heavy_enabled ())
3227 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3228 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3231 /* LOCKING: assumes the GC lock is held */
3233 sgen_restart_world (int generation, GGTimingInfo *timing)
3235 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 sgen_client_restart_world (generation, timing);
3245 world_is_stopped = FALSE;
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, timing, timing ? 2 : 0);
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, NULL);
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 */