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 * This library is free software; you can redistribute it and/or
22 * modify it under the terms of the GNU Library General Public
23 * License 2.0 as published by the Free Software Foundation;
25 * This library is distributed in the hope that it will be useful,
26 * but WITHOUT ANY WARRANTY; without even the implied warranty of
27 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
28 * Library General Public License for more details.
30 * You should have received a copy of the GNU Library General Public
31 * License 2.0 along with this library; if not, write to the Free
32 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
34 * Important: allocation provides always zeroed memory, having to do
35 * a memset after allocation is deadly for performance.
36 * Memory usage at startup is currently as follows:
38 * 64 KB internal space
40 * We should provide a small memory config with half the sizes
42 * We currently try to make as few mono assumptions as possible:
43 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
45 * 2) gc descriptor is the second word in the vtable (first word in the class)
46 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
47 * 4) there is a function to get an object's size and the number of
48 * elements in an array.
49 * 5) we know the special way bounds are allocated for complex arrays
50 * 6) we know about proxies and how to treat them when domains are unloaded
52 * Always try to keep stack usage to a minimum: no recursive behaviour
53 * and no large stack allocs.
55 * General description.
56 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
57 * When the nursery is full we start a nursery collection: this is performed with a
59 * When the old generation is full we start a copying GC of the old generation as well:
60 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
61 * in the future. Maybe we'll even do both during the same collection like IMMIX.
63 * The things that complicate this description are:
64 * *) pinned objects: we can't move them so we need to keep track of them
65 * *) no precise info of the thread stacks and registers: we need to be able to
66 * quickly find the objects that may be referenced conservatively and pin them
67 * (this makes the first issues more important)
68 * *) large objects are too expensive to be dealt with using copying GC: we handle them
69 * with mark/sweep during major collections
70 * *) some objects need to not move even if they are small (interned strings, Type handles):
71 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
72 * PinnedChunks regions
78 *) we could have a function pointer in MonoClass to implement
79 customized write barriers for value types
81 *) investigate the stuff needed to advance a thread to a GC-safe
82 point (single-stepping, read from unmapped memory etc) and implement it.
83 This would enable us to inline allocations and write barriers, for example,
84 or at least parts of them, like the write barrier checks.
85 We may need this also for handling precise info on stacks, even simple things
86 as having uninitialized data on the stack and having to wait for the prolog
87 to zero it. Not an issue for the last frame that we scan conservatively.
88 We could always not trust the value in the slots anyway.
90 *) modify the jit to save info about references in stack locations:
91 this can be done just for locals as a start, so that at least
92 part of the stack is handled precisely.
94 *) test/fix endianess issues
96 *) Implement a card table as the write barrier instead of remembered
97 sets? Card tables are not easy to implement with our current
98 memory layout. We have several different kinds of major heap
99 objects: Small objects in regular blocks, small objects in pinned
100 chunks and LOS objects. If we just have a pointer we have no way
101 to tell which kind of object it points into, therefore we cannot
102 know where its card table is. The least we have to do to make
103 this happen is to get rid of write barriers for indirect stores.
106 *) Get rid of write barriers for indirect stores. We can do this by
107 telling the GC to wbarrier-register an object once we do an ldloca
108 or ldelema on it, and to unregister it once it's not used anymore
109 (it can only travel downwards on the stack). The problem with
110 unregistering is that it needs to happen eventually no matter
111 what, even if exceptions are thrown, the thread aborts, etc.
112 Rodrigo suggested that we could do only the registering part and
113 let the collector find out (pessimistically) when it's safe to
114 unregister, namely when the stack pointer of the thread that
115 registered the object is higher than it was when the registering
116 happened. This might make for a good first implementation to get
117 some data on performance.
119 *) Some sort of blacklist support? Blacklists is a concept from the
120 Boehm GC: if during a conservative scan we find pointers to an
121 area which we might use as heap, we mark that area as unusable, so
122 pointer retention by random pinning pointers is reduced.
124 *) experiment with max small object size (very small right now - 2kb,
125 because it's tied to the max freelist size)
127 *) add an option to mmap the whole heap in one chunk: it makes for many
128 simplifications in the checks (put the nursery at the top and just use a single
129 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
130 not flexible (too much of the address space may be used by default or we can't
131 increase the heap as needed) and we'd need a race-free mechanism to return memory
132 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
133 was written to, munmap is needed, but the following mmap may not find the same segment
136 *) memzero the major fragments after restarting the world and optionally a smaller
139 *) investigate having fragment zeroing threads
141 *) separate locks for finalization and other minor stuff to reduce
144 *) try a different copying order to improve memory locality
146 *) a thread abort after a store but before the write barrier will
147 prevent the write barrier from executing
149 *) specialized dynamically generated markers/copiers
151 *) Dynamically adjust TLAB size to the number of threads. If we have
152 too many threads that do allocation, we might need smaller TLABs,
153 and we might get better performance with larger TLABs if we only
154 have a handful of threads. We could sum up the space left in all
155 assigned TLABs and if that's more than some percentage of the
156 nursery size, reduce the TLAB size.
158 *) Explore placing unreachable objects on unused nursery memory.
159 Instead of memset'ng a region to zero, place an int[] covering it.
160 A good place to start is add_nursery_frag. The tricky thing here is
161 placing those objects atomically outside of a collection.
163 *) Allocation should use asymmetric Dekker synchronization:
164 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
165 This should help weak consistency archs.
172 #define _XOPEN_SOURCE
173 #define _DARWIN_C_SOURCE
179 #ifdef HAVE_PTHREAD_H
182 #ifdef HAVE_PTHREAD_NP_H
183 #include <pthread_np.h>
191 #include "mono/sgen/sgen-gc.h"
192 #include "mono/sgen/sgen-cardtable.h"
193 #include "mono/sgen/sgen-protocol.h"
194 #include "mono/sgen/sgen-memory-governor.h"
195 #include "mono/sgen/sgen-hash-table.h"
196 #include "mono/sgen/sgen-cardtable.h"
197 #include "mono/sgen/sgen-pinning.h"
198 #include "mono/sgen/sgen-workers.h"
199 #include "mono/sgen/sgen-client.h"
200 #include "mono/sgen/sgen-pointer-queue.h"
201 #include "mono/sgen/gc-internal-agnostic.h"
202 #include "mono/utils/mono-proclib.h"
203 #include "mono/utils/mono-memory-model.h"
204 #include "mono/utils/hazard-pointer.h"
206 #include <mono/utils/memcheck.h>
208 #undef pthread_create
210 #undef pthread_detach
213 * ######################################################################
214 * ######## Types and constants used by the GC.
215 * ######################################################################
218 /* 0 means not initialized, 1 is initialized, -1 means in progress */
219 static int gc_initialized = 0;
220 /* If set, check if we need to do something every X allocations */
221 gboolean has_per_allocation_action;
222 /* If set, do a heap check every X allocation */
223 guint32 verify_before_allocs = 0;
224 /* If set, do a minor collection before every X allocation */
225 guint32 collect_before_allocs = 0;
226 /* If set, do a whole heap check before each collection */
227 static gboolean whole_heap_check_before_collection = FALSE;
228 /* If set, do a heap consistency check before each minor collection */
229 static gboolean consistency_check_at_minor_collection = FALSE;
230 /* If set, do a mod union consistency check before each finishing collection pause */
231 static gboolean mod_union_consistency_check = FALSE;
232 /* If set, check whether mark bits are consistent after major collections */
233 static gboolean check_mark_bits_after_major_collection = FALSE;
234 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
235 static gboolean check_nursery_objects_pinned = FALSE;
236 /* If set, do a few checks when the concurrent collector is used */
237 static gboolean do_concurrent_checks = FALSE;
238 /* If set, do a plausibility check on the scan_starts before and after
240 static gboolean do_scan_starts_check = FALSE;
242 static gboolean disable_minor_collections = FALSE;
243 static gboolean disable_major_collections = FALSE;
244 static gboolean do_verify_nursery = FALSE;
245 static gboolean do_dump_nursery_content = FALSE;
246 static gboolean enable_nursery_canaries = FALSE;
248 static gboolean precleaning_enabled = TRUE;
250 #ifdef HEAVY_STATISTICS
251 guint64 stat_objects_alloced_degraded = 0;
252 guint64 stat_bytes_alloced_degraded = 0;
254 guint64 stat_copy_object_called_nursery = 0;
255 guint64 stat_objects_copied_nursery = 0;
256 guint64 stat_copy_object_called_major = 0;
257 guint64 stat_objects_copied_major = 0;
259 guint64 stat_scan_object_called_nursery = 0;
260 guint64 stat_scan_object_called_major = 0;
262 guint64 stat_slots_allocated_in_vain;
264 guint64 stat_nursery_copy_object_failed_from_space = 0;
265 guint64 stat_nursery_copy_object_failed_forwarded = 0;
266 guint64 stat_nursery_copy_object_failed_pinned = 0;
267 guint64 stat_nursery_copy_object_failed_to_space = 0;
269 static guint64 stat_wbarrier_add_to_global_remset = 0;
270 static guint64 stat_wbarrier_arrayref_copy = 0;
271 static guint64 stat_wbarrier_generic_store = 0;
272 static guint64 stat_wbarrier_generic_store_atomic = 0;
273 static guint64 stat_wbarrier_set_root = 0;
276 static guint64 stat_pinned_objects = 0;
278 static guint64 time_minor_pre_collection_fragment_clear = 0;
279 static guint64 time_minor_pinning = 0;
280 static guint64 time_minor_scan_remsets = 0;
281 static guint64 time_minor_scan_pinned = 0;
282 static guint64 time_minor_scan_roots = 0;
283 static guint64 time_minor_finish_gray_stack = 0;
284 static guint64 time_minor_fragment_creation = 0;
286 static guint64 time_major_pre_collection_fragment_clear = 0;
287 static guint64 time_major_pinning = 0;
288 static guint64 time_major_scan_pinned = 0;
289 static guint64 time_major_scan_roots = 0;
290 static guint64 time_major_scan_mod_union = 0;
291 static guint64 time_major_finish_gray_stack = 0;
292 static guint64 time_major_free_bigobjs = 0;
293 static guint64 time_major_los_sweep = 0;
294 static guint64 time_major_sweep = 0;
295 static guint64 time_major_fragment_creation = 0;
297 static guint64 time_max = 0;
299 static SGEN_TV_DECLARE (time_major_conc_collection_start);
300 static SGEN_TV_DECLARE (time_major_conc_collection_end);
302 static SGEN_TV_DECLARE (last_minor_collection_start_tv);
303 static SGEN_TV_DECLARE (last_minor_collection_end_tv);
305 int gc_debug_level = 0;
310 mono_gc_flush_info (void)
312 fflush (gc_debug_file);
316 #define TV_DECLARE SGEN_TV_DECLARE
317 #define TV_GETTIME SGEN_TV_GETTIME
318 #define TV_ELAPSED SGEN_TV_ELAPSED
320 static SGEN_TV_DECLARE (sgen_init_timestamp);
322 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
324 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
325 #define object_is_pinned SGEN_OBJECT_IS_PINNED
326 #define pin_object SGEN_PIN_OBJECT
328 #define ptr_in_nursery sgen_ptr_in_nursery
330 #define LOAD_VTABLE SGEN_LOAD_VTABLE
333 nursery_canaries_enabled (void)
335 return enable_nursery_canaries;
338 #define safe_object_get_size sgen_safe_object_get_size
341 * ######################################################################
342 * ######## Global data.
343 * ######################################################################
345 MonoCoopMutex gc_mutex;
347 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
349 size_t degraded_mode = 0;
351 static mword bytes_pinned_from_failed_allocation = 0;
353 GCMemSection *nursery_section = NULL;
354 static volatile mword lowest_heap_address = ~(mword)0;
355 static volatile mword highest_heap_address = 0;
357 MonoCoopMutex sgen_interruption_mutex;
359 int current_collection_generation = -1;
360 static volatile gboolean concurrent_collection_in_progress = FALSE;
362 /* objects that are ready to be finalized */
363 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
364 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
366 /* registered roots: the key to the hash is the root start address */
368 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
370 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
371 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
372 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
373 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
375 static mword roots_size = 0; /* amount of memory in the root set */
377 /* The size of a TLAB */
378 /* The bigger the value, the less often we have to go to the slow path to allocate a new
379 * one, but the more space is wasted by threads not allocating much memory.
381 * FIXME: Make this self-tuning for each thread.
383 guint32 tlab_size = (1024 * 4);
385 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
387 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
389 #define ALIGN_UP SGEN_ALIGN_UP
391 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
392 MonoNativeThreadId main_gc_thread = NULL;
395 /*Object was pinned during the current collection*/
396 static mword objects_pinned;
399 * ######################################################################
400 * ######## Macros and function declarations.
401 * ######################################################################
404 typedef SgenGrayQueue GrayQueue;
406 /* forward declarations */
407 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
409 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
410 static void finish_gray_stack (int generation, ScanCopyContext ctx);
413 SgenMajorCollector major_collector;
414 SgenMinorCollector sgen_minor_collector;
415 /* FIXME: get rid of this */
416 static GrayQueue gray_queue;
418 static SgenRememberedSet remset;
420 /* The gray queue to use from the main collection thread. */
421 #define WORKERS_DISTRIBUTE_GRAY_QUEUE (&gray_queue)
424 * The gray queue a worker job must use. If we're not parallel or
425 * concurrent, we use the main gray queue.
427 static SgenGrayQueue*
428 sgen_workers_get_job_gray_queue (WorkerData *worker_data)
430 return worker_data ? &worker_data->private_gray_queue : WORKERS_DISTRIBUTE_GRAY_QUEUE;
434 gray_queue_redirect (SgenGrayQueue *queue)
436 gboolean wake = FALSE;
439 GrayQueueSection *section = sgen_gray_object_dequeue_section (queue);
442 sgen_section_gray_queue_enqueue ((SgenSectionGrayQueue *)queue->alloc_prepare_data, section);
447 g_assert (concurrent_collection_in_progress);
448 sgen_workers_ensure_awake ();
453 gray_queue_enable_redirect (SgenGrayQueue *queue)
455 if (!concurrent_collection_in_progress)
458 sgen_gray_queue_set_alloc_prepare (queue, gray_queue_redirect, sgen_workers_get_distribute_section_gray_queue ());
459 gray_queue_redirect (queue);
463 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
465 while (start < end) {
469 if (!*(void**)start) {
470 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
475 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
481 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
482 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
483 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
484 callback ((GCObject*)obj, size, data);
485 CANARIFY_SIZE (size);
487 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
495 * sgen_add_to_global_remset:
497 * The global remset contains locations which point into newspace after
498 * a minor collection. This can happen if the objects they point to are pinned.
500 * LOCKING: If called from a parallel collector, the global remset
501 * lock must be held. For serial collectors that is not necessary.
504 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
506 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
508 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
510 if (!major_collector.is_concurrent) {
511 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
513 if (current_collection_generation == -1)
514 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
517 if (!object_is_pinned (obj))
518 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");
519 else if (sgen_cement_lookup_or_register (obj))
522 remset.record_pointer (ptr);
524 sgen_pin_stats_register_global_remset (obj);
526 SGEN_LOG (8, "Adding global remset for %p", ptr);
527 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
531 * sgen_drain_gray_stack:
533 * Scan objects in the gray stack until the stack is empty. This should be called
534 * frequently after each object is copied, to achieve better locality and cache
539 sgen_drain_gray_stack (ScanCopyContext ctx)
541 ScanObjectFunc scan_func = ctx.ops->scan_object;
542 GrayQueue *queue = ctx.queue;
544 if (ctx.ops->drain_gray_stack)
545 return ctx.ops->drain_gray_stack (queue);
550 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
553 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
554 scan_func (obj, desc, queue);
560 * Addresses in the pin queue are already sorted. This function finds
561 * the object header for each address and pins the object. The
562 * addresses must be inside the nursery section. The (start of the)
563 * address array is overwritten with the addresses of the actually
564 * pinned objects. Return the number of pinned objects.
567 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
569 GCMemSection *section = nursery_section;
570 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
571 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
572 void *start_nursery = section->data;
573 void *end_nursery = section->next_data;
578 void *pinning_front = start_nursery;
580 void **definitely_pinned = start;
581 ScanObjectFunc scan_func = ctx.ops->scan_object;
582 SgenGrayQueue *queue = ctx.queue;
584 sgen_nursery_allocator_prepare_for_pinning ();
586 while (start < end) {
587 GCObject *obj_to_pin = NULL;
588 size_t obj_to_pin_size = 0;
593 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
594 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
601 SGEN_LOG (5, "Considering pinning addr %p", addr);
602 /* We've already processed everything up to pinning_front. */
603 if (addr < pinning_front) {
609 * Find the closest scan start <= addr. We might search backward in the
610 * scan_starts array because entries might be NULL. In the worst case we
611 * start at start_nursery.
613 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
614 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
615 search_start = (void*)section->scan_starts [idx];
616 if (!search_start || search_start > addr) {
619 search_start = section->scan_starts [idx];
620 if (search_start && search_start <= addr)
623 if (!search_start || search_start > addr)
624 search_start = start_nursery;
628 * If the pinning front is closer than the scan start we found, start
629 * searching at the front.
631 if (search_start < pinning_front)
632 search_start = pinning_front;
635 * Now addr should be in an object a short distance from search_start.
637 * search_start must point to zeroed mem or point to an object.
640 size_t obj_size, canarified_obj_size;
643 if (!*(void**)search_start) {
644 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
645 /* The loop condition makes sure we don't overrun addr. */
649 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
652 * Filler arrays are marked by an invalid sync word. We don't
653 * consider them for pinning. They are not delimited by canaries,
656 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
657 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
658 CANARIFY_SIZE (canarified_obj_size);
660 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
661 /* This is the object we're looking for. */
662 obj_to_pin = (GCObject*)search_start;
663 obj_to_pin_size = canarified_obj_size;
668 /* Skip to the next object */
669 search_start = (void*)((char*)search_start + canarified_obj_size);
670 } while (search_start <= addr);
672 /* We've searched past the address we were looking for. */
674 pinning_front = search_start;
675 goto next_pin_queue_entry;
679 * We've found an object to pin. It might still be a dummy array, but we
680 * can advance the pinning front in any case.
682 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
685 * If this is a dummy array marking the beginning of a nursery
686 * fragment, we don't pin it.
688 if (sgen_client_object_is_array_fill (obj_to_pin))
689 goto next_pin_queue_entry;
692 * Finally - pin the object!
694 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
695 if (do_scan_objects) {
696 scan_func (obj_to_pin, desc, queue);
698 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
699 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
700 binary_protocol_pin (obj_to_pin,
701 (gpointer)LOAD_VTABLE (obj_to_pin),
702 safe_object_get_size (obj_to_pin));
704 pin_object (obj_to_pin);
705 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
706 sgen_pin_stats_register_object (obj_to_pin, obj_to_pin_size);
707 definitely_pinned [count] = obj_to_pin;
711 next_pin_queue_entry:
715 sgen_client_nursery_objects_pinned (definitely_pinned, count);
716 stat_pinned_objects += count;
721 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
725 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
728 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
729 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
733 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
734 * when we can't promote an object because we're out of memory.
737 sgen_pin_object (GCObject *object, GrayQueue *queue)
740 * All pinned objects are assumed to have been staged, so we need to stage as well.
741 * Also, the count of staged objects shows that "late pinning" happened.
743 sgen_pin_stage_ptr (object);
745 SGEN_PIN_OBJECT (object);
746 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
749 sgen_pin_stats_register_object (object, safe_object_get_size (object));
751 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
754 /* Sort the addresses in array in increasing order.
755 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
758 sgen_sort_addresses (void **array, size_t size)
763 for (i = 1; i < size; ++i) {
766 size_t parent = (child - 1) / 2;
768 if (array [parent] >= array [child])
771 tmp = array [parent];
772 array [parent] = array [child];
779 for (i = size - 1; i > 0; --i) {
782 array [i] = array [0];
788 while (root * 2 + 1 <= end) {
789 size_t child = root * 2 + 1;
791 if (child < end && array [child] < array [child + 1])
793 if (array [root] >= array [child])
797 array [root] = array [child];
806 * Scan the memory between start and end and queue values which could be pointers
807 * to the area between start_nursery and end_nursery for later consideration.
808 * Typically used for thread stacks.
811 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
815 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
817 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
818 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
821 while (start < end) {
823 * *start can point to the middle of an object
824 * note: should we handle pointing at the end of an object?
825 * pinning in C# code disallows pointing at the end of an object
826 * but there is some small chance that an optimizing C compiler
827 * may keep the only reference to an object by pointing
828 * at the end of it. We ignore this small chance for now.
829 * Pointers to the end of an object are indistinguishable
830 * from pointers to the start of the next object in memory
831 * so if we allow that we'd need to pin two objects...
832 * We queue the pointer in an array, the
833 * array will then be sorted and uniqued. This way
834 * we can coalesce several pinning pointers and it should
835 * be faster since we'd do a memory scan with increasing
836 * addresses. Note: we can align the address to the allocation
837 * alignment, so the unique process is more effective.
839 mword addr = (mword)*start;
840 addr &= ~(ALLOC_ALIGN - 1);
841 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
842 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
843 sgen_pin_stage_ptr ((void*)addr);
844 binary_protocol_pin_stage (start, (void*)addr);
845 sgen_pin_stats_register_address ((char*)addr, pin_type);
851 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
855 * The first thing we do in a collection is to identify pinned objects.
856 * This function considers all the areas of memory that need to be
857 * conservatively scanned.
860 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
864 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);
865 /* objects pinned from the API are inside these roots */
866 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
867 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
868 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
869 } SGEN_HASH_TABLE_FOREACH_END;
870 /* now deal with the thread stacks
871 * in the future we should be able to conservatively scan only:
872 * *) the cpu registers
873 * *) the unmanaged stack frames
874 * *) the _last_ managed stack frame
875 * *) pointers slots in managed frames
877 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
881 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
883 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
884 ctx->ops->copy_or_mark_object (obj, ctx->queue);
888 * The memory area from start_root to end_root contains pointers to objects.
889 * Their position is precisely described by @desc (this means that the pointer
890 * can be either NULL or the pointer to the start of an object).
891 * This functions copies them to to_space updates them.
893 * This function is not thread-safe!
896 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
898 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
899 SgenGrayQueue *queue = ctx.queue;
901 switch (desc & ROOT_DESC_TYPE_MASK) {
902 case ROOT_DESC_BITMAP:
903 desc >>= ROOT_DESC_TYPE_SHIFT;
905 if ((desc & 1) && *start_root) {
906 copy_func ((GCObject**)start_root, queue);
907 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
913 case ROOT_DESC_COMPLEX: {
914 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
915 gsize bwords = (*bitmap_data) - 1;
916 void **start_run = start_root;
918 while (bwords-- > 0) {
919 gsize bmap = *bitmap_data++;
920 void **objptr = start_run;
922 if ((bmap & 1) && *objptr) {
923 copy_func ((GCObject**)objptr, queue);
924 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
929 start_run += GC_BITS_PER_WORD;
933 case ROOT_DESC_USER: {
934 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
935 marker (start_root, single_arg_user_copy_or_mark, &ctx);
938 case ROOT_DESC_RUN_LEN:
939 g_assert_not_reached ();
941 g_assert_not_reached ();
946 reset_heap_boundaries (void)
948 lowest_heap_address = ~(mword)0;
949 highest_heap_address = 0;
953 sgen_update_heap_boundaries (mword low, mword high)
958 old = lowest_heap_address;
961 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
964 old = highest_heap_address;
967 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
971 * Allocate and setup the data structures needed to be able to allocate objects
972 * in the nursery. The nursery is stored in nursery_section.
977 GCMemSection *section;
984 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
985 /* later we will alloc a larger area for the nursery but only activate
986 * what we need. The rest will be used as expansion if we have too many pinned
987 * objects in the existing nursery.
989 /* FIXME: handle OOM */
990 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
992 alloc_size = sgen_nursery_size;
994 /* If there isn't enough space even for the nursery we should simply abort. */
995 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
997 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
998 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
999 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 ());
1000 section->data = section->next_data = data;
1001 section->size = alloc_size;
1002 section->end_data = data + sgen_nursery_size;
1003 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
1004 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
1005 section->num_scan_start = scan_starts;
1007 nursery_section = section;
1009 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
1013 mono_gc_get_logfile (void)
1015 return gc_debug_file;
1019 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1021 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1022 SgenGrayQueue *queue = ctx.queue;
1025 for (i = 0; i < fin_queue->next_slot; ++i) {
1026 GCObject *obj = (GCObject *)fin_queue->data [i];
1029 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1030 copy_func ((GCObject**)&fin_queue->data [i], queue);
1035 generation_name (int generation)
1037 switch (generation) {
1038 case GENERATION_NURSERY: return "nursery";
1039 case GENERATION_OLD: return "old";
1040 default: g_assert_not_reached ();
1045 sgen_generation_name (int generation)
1047 return generation_name (generation);
1051 finish_gray_stack (int generation, ScanCopyContext ctx)
1055 int done_with_ephemerons, ephemeron_rounds = 0;
1056 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1057 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1058 SgenGrayQueue *queue = ctx.queue;
1060 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1062 * We copied all the reachable objects. Now it's the time to copy
1063 * the objects that were not referenced by the roots, but by the copied objects.
1064 * we built a stack of objects pointed to by gray_start: they are
1065 * additional roots and we may add more items as we go.
1066 * We loop until gray_start == gray_objects which means no more objects have
1067 * been added. Note this is iterative: no recursion is involved.
1068 * We need to walk the LO list as well in search of marked big objects
1069 * (use a flag since this is needed only on major collections). We need to loop
1070 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1071 * To achieve better cache locality and cache usage, we drain the gray stack
1072 * frequently, after each object is copied, and just finish the work here.
1074 sgen_drain_gray_stack (ctx);
1076 SGEN_LOG (2, "%s generation done", generation_name (generation));
1079 Reset bridge data, we might have lingering data from a previous collection if this is a major
1080 collection trigged by minor overflow.
1082 We must reset the gathered bridges since their original block might be evacuated due to major
1083 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1085 if (sgen_client_bridge_need_processing ())
1086 sgen_client_bridge_reset_data ();
1089 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1090 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1091 * objects that are in fact reachable.
1093 done_with_ephemerons = 0;
1095 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1096 sgen_drain_gray_stack (ctx);
1098 } while (!done_with_ephemerons);
1100 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1102 if (sgen_client_bridge_need_processing ()) {
1103 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1104 sgen_drain_gray_stack (ctx);
1105 sgen_collect_bridge_objects (generation, ctx);
1106 if (generation == GENERATION_OLD)
1107 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1110 Do the first bridge step here, as the collector liveness state will become useless after that.
1112 An important optimization is to only proccess the possibly dead part of the object graph and skip
1113 over all live objects as we transitively know everything they point must be alive too.
1115 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1117 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1118 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1121 sgen_client_bridge_processing_stw_step ();
1125 Make sure we drain the gray stack before processing disappearing links and finalizers.
1126 If we don't make sure it is empty we might wrongly see a live object as dead.
1128 sgen_drain_gray_stack (ctx);
1131 We must clear weak links that don't track resurrection before processing object ready for
1132 finalization so they can be cleared before that.
1134 sgen_null_link_in_range (generation, ctx, FALSE);
1135 if (generation == GENERATION_OLD)
1136 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1139 /* walk the finalization queue and move also the objects that need to be
1140 * finalized: use the finalized objects as new roots so the objects they depend
1141 * on are also not reclaimed. As with the roots above, only objects in the nursery
1142 * are marked/copied.
1144 sgen_finalize_in_range (generation, ctx);
1145 if (generation == GENERATION_OLD)
1146 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1147 /* drain the new stack that might have been created */
1148 SGEN_LOG (6, "Precise scan of gray area post fin");
1149 sgen_drain_gray_stack (ctx);
1152 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1154 done_with_ephemerons = 0;
1156 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1157 sgen_drain_gray_stack (ctx);
1159 } while (!done_with_ephemerons);
1161 sgen_client_clear_unreachable_ephemerons (ctx);
1164 * We clear togglerefs only after all possible chances of revival are done.
1165 * This is semantically more inline with what users expect and it allows for
1166 * user finalizers to correctly interact with TR objects.
1168 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1171 SGEN_LOG (2, "Finalize queue handling scan for %s generation: %lld usecs %d ephemeron rounds", generation_name (generation), TV_ELAPSED (atv, btv), ephemeron_rounds);
1174 * handle disappearing links
1175 * Note we do this after checking the finalization queue because if an object
1176 * survives (at least long enough to be finalized) we don't clear the link.
1177 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1178 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1181 g_assert (sgen_gray_object_queue_is_empty (queue));
1183 sgen_null_link_in_range (generation, ctx, TRUE);
1184 if (generation == GENERATION_OLD)
1185 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1186 if (sgen_gray_object_queue_is_empty (queue))
1188 sgen_drain_gray_stack (ctx);
1191 g_assert (sgen_gray_object_queue_is_empty (queue));
1193 sgen_gray_object_queue_trim_free_list (queue);
1194 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1198 sgen_check_section_scan_starts (GCMemSection *section)
1201 for (i = 0; i < section->num_scan_start; ++i) {
1202 if (section->scan_starts [i]) {
1203 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1204 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1210 check_scan_starts (void)
1212 if (!do_scan_starts_check)
1214 sgen_check_section_scan_starts (nursery_section);
1215 major_collector.check_scan_starts ();
1219 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1223 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1224 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1225 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1226 } SGEN_HASH_TABLE_FOREACH_END;
1232 static gboolean inited = FALSE;
1237 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1239 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1240 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1241 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1242 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1243 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1244 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1246 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1247 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1248 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1249 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1250 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1251 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1252 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1253 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1254 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1255 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1257 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1259 #ifdef HEAVY_STATISTICS
1260 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1261 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1262 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1263 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1264 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1266 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1267 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1269 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1270 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1271 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1272 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1274 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1275 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1277 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1279 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1280 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1281 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1282 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1284 sgen_nursery_allocator_init_heavy_stats ();
1292 reset_pinned_from_failed_allocation (void)
1294 bytes_pinned_from_failed_allocation = 0;
1298 sgen_set_pinned_from_failed_allocation (mword objsize)
1300 bytes_pinned_from_failed_allocation += objsize;
1304 sgen_collection_is_concurrent (void)
1306 switch (current_collection_generation) {
1307 case GENERATION_NURSERY:
1309 case GENERATION_OLD:
1310 return concurrent_collection_in_progress;
1312 g_error ("Invalid current generation %d", current_collection_generation);
1318 sgen_concurrent_collection_in_progress (void)
1320 return concurrent_collection_in_progress;
1324 SgenThreadPoolJob job;
1325 SgenObjectOperations *ops;
1329 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1331 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1332 ScanJob *job_data = (ScanJob*)job;
1333 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1334 remset.scan_remsets (ctx);
1338 SgenThreadPoolJob job;
1339 SgenObjectOperations *ops;
1343 } ScanFromRegisteredRootsJob;
1346 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1348 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1349 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1350 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1352 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1356 SgenThreadPoolJob job;
1357 SgenObjectOperations *ops;
1360 } ScanThreadDataJob;
1363 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1365 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1366 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1367 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1369 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1373 SgenThreadPoolJob job;
1374 SgenObjectOperations *ops;
1375 SgenPointerQueue *queue;
1376 } ScanFinalizerEntriesJob;
1379 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1381 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1382 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1383 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1385 scan_finalizer_entries (job_data->queue, ctx);
1389 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1391 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1392 ScanJob *job_data = (ScanJob*)job;
1393 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1395 g_assert (concurrent_collection_in_progress);
1396 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1400 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1402 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1403 ScanJob *job_data = (ScanJob*)job;
1404 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1406 g_assert (concurrent_collection_in_progress);
1407 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1411 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1413 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1414 ScanJob *job_data = (ScanJob*)job;
1415 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1417 g_assert (concurrent_collection_in_progress);
1419 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1420 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1424 init_gray_queue (gboolean use_workers)
1427 sgen_workers_init_distribute_gray_queue ();
1428 sgen_gray_object_queue_init (&gray_queue, NULL);
1432 enqueue_scan_from_roots_jobs (char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1434 ScanFromRegisteredRootsJob *scrrj;
1435 ScanThreadDataJob *stdj;
1436 ScanFinalizerEntriesJob *sfej;
1438 /* registered roots, this includes static fields */
1440 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", 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_NORMAL;
1445 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1447 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1449 scrrj->heap_start = heap_start;
1450 scrrj->heap_end = heap_end;
1451 scrrj->root_type = ROOT_TYPE_WBARRIER;
1452 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1456 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1457 stdj->heap_start = heap_start;
1458 stdj->heap_end = heap_end;
1459 sgen_workers_enqueue_job (&stdj->job, enqueue);
1461 /* Scan the list of objects ready for finalization. */
1463 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1464 sfej->queue = &fin_ready_queue;
1466 sgen_workers_enqueue_job (&sfej->job, enqueue);
1468 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1469 sfej->queue = &critical_fin_queue;
1471 sgen_workers_enqueue_job (&sfej->job, enqueue);
1475 * Perform a nursery collection.
1477 * Return whether any objects were late-pinned due to being out of memory.
1480 collect_nursery (SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1482 gboolean needs_major;
1483 size_t max_garbage_amount;
1485 mword fragment_total;
1487 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1488 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue);
1492 if (disable_minor_collections)
1495 TV_GETTIME (last_minor_collection_start_tv);
1496 atv = last_minor_collection_start_tv;
1498 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1500 if (do_verify_nursery || do_dump_nursery_content)
1501 sgen_debug_verify_nursery (do_dump_nursery_content);
1503 current_collection_generation = GENERATION_NURSERY;
1505 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1507 reset_pinned_from_failed_allocation ();
1509 check_scan_starts ();
1511 sgen_nursery_alloc_prepare_for_minor ();
1515 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1516 /* FIXME: optimize later to use the higher address where an object can be present */
1517 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1519 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 ()));
1520 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1521 g_assert (nursery_section->size >= max_garbage_amount);
1523 /* world must be stopped already */
1525 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1527 sgen_client_pre_collection_checks ();
1529 nursery_section->next_data = nursery_next;
1531 major_collector.start_nursery_collection ();
1533 sgen_memgov_minor_collection_start ();
1535 init_gray_queue (FALSE);
1537 gc_stats.minor_gc_count ++;
1539 if (whole_heap_check_before_collection) {
1540 sgen_clear_nursery_fragments ();
1541 sgen_check_whole_heap (finish_up_concurrent_mark);
1543 if (consistency_check_at_minor_collection)
1544 sgen_check_consistency ();
1546 sgen_process_fin_stage_entries ();
1548 /* pin from pinned handles */
1549 sgen_init_pinning ();
1550 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1551 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1552 /* pin cemented objects */
1553 sgen_pin_cemented_objects ();
1554 /* identify pinned objects */
1555 sgen_optimize_pin_queue ();
1556 sgen_pinning_setup_section (nursery_section);
1558 pin_objects_in_nursery (FALSE, ctx);
1559 sgen_pinning_trim_queue_to_section (nursery_section);
1562 time_minor_pinning += TV_ELAPSED (btv, atv);
1563 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), TV_ELAPSED (btv, atv));
1564 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1566 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1567 sj->ops = object_ops;
1568 sgen_workers_enqueue_job (&sj->job, FALSE);
1570 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1572 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1573 SGEN_LOG (2, "Old generation scan: %lld usecs", TV_ELAPSED (atv, btv));
1575 sgen_pin_stats_print_class_stats ();
1577 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1578 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1581 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1583 enqueue_scan_from_roots_jobs (sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1586 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1588 finish_gray_stack (GENERATION_NURSERY, ctx);
1591 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1592 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1594 if (objects_pinned) {
1595 sgen_optimize_pin_queue ();
1596 sgen_pinning_setup_section (nursery_section);
1599 /* walk the pin_queue, build up the fragment list of free memory, unmark
1600 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1603 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1604 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1605 if (!fragment_total)
1608 /* Clear TLABs for all threads */
1609 sgen_clear_tlabs ();
1611 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1613 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1614 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1616 if (consistency_check_at_minor_collection)
1617 sgen_check_major_refs ();
1619 major_collector.finish_nursery_collection ();
1621 TV_GETTIME (last_minor_collection_end_tv);
1622 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1624 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1626 /* prepare the pin queue for the next collection */
1627 sgen_finish_pinning ();
1628 if (sgen_have_pending_finalizers ()) {
1629 SGEN_LOG (4, "Finalizer-thread wakeup");
1630 sgen_client_finalize_notify ();
1632 sgen_pin_stats_reset ();
1633 /* clear cemented hash */
1634 sgen_cement_clear_below_threshold ();
1636 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
1638 remset.finish_minor_collection ();
1640 check_scan_starts ();
1642 binary_protocol_flush_buffers (FALSE);
1644 sgen_memgov_minor_collection_end ();
1646 /*objects are late pinned because of lack of memory, so a major is a good call*/
1647 needs_major = objects_pinned > 0;
1648 current_collection_generation = -1;
1651 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1653 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1654 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1660 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1661 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1662 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1663 } CopyOrMarkFromRootsMode;
1666 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1671 /* FIXME: only use these values for the precise scan
1672 * note that to_space pointers should be excluded anyway...
1674 char *heap_start = NULL;
1675 char *heap_end = (char*)-1;
1676 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, WORKERS_DISTRIBUTE_GRAY_QUEUE);
1677 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1679 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1681 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1682 /*This cleans up unused fragments */
1683 sgen_nursery_allocator_prepare_for_pinning ();
1685 if (do_concurrent_checks)
1686 sgen_debug_check_nursery_is_clean ();
1688 /* The concurrent collector doesn't touch the nursery. */
1689 sgen_nursery_alloc_prepare_for_major ();
1692 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1696 /* Pinning depends on this */
1697 sgen_clear_nursery_fragments ();
1699 if (whole_heap_check_before_collection)
1700 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1703 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1705 if (!sgen_collection_is_concurrent ())
1706 nursery_section->next_data = sgen_get_nursery_end ();
1707 /* we should also coalesce scanning from sections close to each other
1708 * and deal with pointers outside of the sections later.
1713 sgen_client_pre_collection_checks ();
1716 /* Remsets are not useful for a major collection */
1717 remset.clear_cards ();
1720 sgen_process_fin_stage_entries ();
1723 sgen_init_pinning ();
1724 SGEN_LOG (6, "Collecting pinned addresses");
1725 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1727 sgen_optimize_pin_queue ();
1729 sgen_client_collecting_major_1 ();
1732 * pin_queue now contains all candidate pointers, sorted and
1733 * uniqued. We must do two passes now to figure out which
1734 * objects are pinned.
1736 * The first is to find within the pin_queue the area for each
1737 * section. This requires that the pin_queue be sorted. We
1738 * also process the LOS objects and pinned chunks here.
1740 * The second, destructive, pass is to reduce the section
1741 * areas to pointers to the actually pinned objects.
1743 SGEN_LOG (6, "Pinning from sections");
1744 /* first pass for the sections */
1745 sgen_find_section_pin_queue_start_end (nursery_section);
1746 /* identify possible pointers to the insize of large objects */
1747 SGEN_LOG (6, "Pinning from large objects");
1748 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1750 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1751 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1753 if (sgen_los_object_is_pinned (bigobj->data)) {
1754 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1757 sgen_los_pin_object (bigobj->data);
1758 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1759 GRAY_OBJECT_ENQUEUE (WORKERS_DISTRIBUTE_GRAY_QUEUE, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1760 sgen_pin_stats_register_object (bigobj->data, safe_object_get_size (bigobj->data));
1761 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1762 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1763 (unsigned long)sgen_los_object_size (bigobj));
1765 sgen_client_pinned_los_object (bigobj->data);
1769 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1770 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1771 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1773 major_collector.pin_objects (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1774 if (old_next_pin_slot)
1775 *old_next_pin_slot = sgen_get_pinned_count ();
1778 time_major_pinning += TV_ELAPSED (atv, btv);
1779 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), TV_ELAPSED (atv, btv));
1780 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1782 major_collector.init_to_space ();
1784 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1786 * The concurrent collector doesn't move objects, neither on
1787 * the major heap nor in the nursery, so we can mark even
1788 * before pinning has finished. For the non-concurrent
1789 * collector we start the workers after pinning.
1791 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1792 if (precleaning_enabled) {
1794 /* Mod union preclean job */
1795 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1796 sj->ops = object_ops;
1797 sgen_workers_start_all_workers (object_ops, &sj->job);
1799 sgen_workers_start_all_workers (object_ops, NULL);
1801 gray_queue_enable_redirect (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1802 } else if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1803 if (sgen_workers_have_idle_work ()) {
1804 sgen_workers_start_all_workers (object_ops, NULL);
1805 sgen_workers_join ();
1809 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1810 main_gc_thread = mono_native_thread_self ();
1813 sgen_client_collecting_major_2 ();
1816 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1818 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1821 * FIXME: is this the right context? It doesn't seem to contain a copy function
1822 * unless we're concurrent.
1824 enqueue_scan_from_roots_jobs (heap_start, heap_end, object_ops, mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1827 time_major_scan_roots += TV_ELAPSED (atv, btv);
1829 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1832 /* Mod union card table */
1833 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1834 sj->ops = object_ops;
1835 sgen_workers_enqueue_job (&sj->job, FALSE);
1837 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1838 sj->ops = object_ops;
1839 sgen_workers_enqueue_job (&sj->job, FALSE);
1842 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1845 sgen_pin_stats_print_class_stats ();
1849 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1851 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1853 * Prepare the pin queue for the next collection. Since pinning runs on the worker
1854 * threads we must wait for the jobs to finish before we can reset it.
1856 sgen_workers_wait_for_jobs_finished ();
1857 sgen_finish_pinning ();
1859 sgen_pin_stats_reset ();
1861 if (do_concurrent_checks)
1862 sgen_debug_check_nursery_is_clean ();
1867 major_start_collection (gboolean concurrent, size_t *old_next_pin_slot)
1869 SgenObjectOperations *object_ops;
1871 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1873 current_collection_generation = GENERATION_OLD;
1875 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1878 sgen_cement_reset ();
1881 g_assert (major_collector.is_concurrent);
1882 concurrent_collection_in_progress = TRUE;
1884 object_ops = &major_collector.major_ops_concurrent_start;
1886 object_ops = &major_collector.major_ops_serial;
1889 reset_pinned_from_failed_allocation ();
1891 sgen_memgov_major_collection_start ();
1893 //count_ref_nonref_objs ();
1894 //consistency_check ();
1896 check_scan_starts ();
1899 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1900 gc_stats.major_gc_count ++;
1902 if (major_collector.start_major_collection)
1903 major_collector.start_major_collection ();
1905 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);
1906 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1910 major_finish_collection (const char *reason, size_t old_next_pin_slot, gboolean forced)
1912 ScannedObjectCounts counts;
1913 SgenObjectOperations *object_ops;
1914 mword fragment_total;
1920 if (concurrent_collection_in_progress) {
1921 object_ops = &major_collector.major_ops_concurrent_finish;
1923 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1925 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1927 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1928 main_gc_thread = NULL;
1931 object_ops = &major_collector.major_ops_serial;
1934 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1936 /* all the objects in the heap */
1937 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue));
1939 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1941 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1943 if (objects_pinned) {
1944 g_assert (!concurrent_collection_in_progress);
1947 * This is slow, but we just OOM'd.
1949 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1950 * queue is laid out at this point.
1952 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1954 * We need to reestablish all pinned nursery objects in the pin queue
1955 * because they're needed for fragment creation. Unpinning happens by
1956 * walking the whole queue, so it's not necessary to reestablish where major
1957 * heap block pins are - all we care is that they're still in there
1960 sgen_optimize_pin_queue ();
1961 sgen_find_section_pin_queue_start_end (nursery_section);
1965 reset_heap_boundaries ();
1966 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1968 /* walk the pin_queue, build up the fragment list of free memory, unmark
1969 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1972 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1973 if (!fragment_total)
1975 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1977 if (do_concurrent_checks && concurrent_collection_in_progress)
1978 sgen_debug_check_nursery_is_clean ();
1980 /* prepare the pin queue for the next collection */
1981 sgen_finish_pinning ();
1983 /* Clear TLABs for all threads */
1984 sgen_clear_tlabs ();
1986 sgen_pin_stats_reset ();
1988 sgen_cement_clear_below_threshold ();
1990 if (check_mark_bits_after_major_collection)
1991 sgen_check_heap_marked (concurrent_collection_in_progress);
1994 time_major_fragment_creation += TV_ELAPSED (atv, btv);
1996 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
1997 sgen_memgov_major_pre_sweep ();
2000 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2005 time_major_los_sweep += TV_ELAPSED (atv, btv);
2007 major_collector.sweep ();
2009 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2012 time_major_sweep += TV_ELAPSED (btv, atv);
2014 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2016 if (sgen_have_pending_finalizers ()) {
2017 SGEN_LOG (4, "Finalizer-thread wakeup");
2018 sgen_client_finalize_notify ();
2021 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2023 sgen_memgov_major_collection_end (forced);
2024 current_collection_generation = -1;
2026 memset (&counts, 0, sizeof (ScannedObjectCounts));
2027 major_collector.finish_major_collection (&counts);
2029 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
2031 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2032 if (concurrent_collection_in_progress)
2033 concurrent_collection_in_progress = FALSE;
2035 check_scan_starts ();
2037 binary_protocol_flush_buffers (FALSE);
2039 //consistency_check ();
2041 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2045 major_do_collection (const char *reason, gboolean forced)
2047 TV_DECLARE (time_start);
2048 TV_DECLARE (time_end);
2049 size_t old_next_pin_slot;
2051 if (disable_major_collections)
2054 if (major_collector.get_and_reset_num_major_objects_marked) {
2055 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2056 g_assert (!num_marked);
2059 /* world must be stopped already */
2060 TV_GETTIME (time_start);
2062 major_start_collection (FALSE, &old_next_pin_slot);
2063 major_finish_collection (reason, old_next_pin_slot, forced);
2065 TV_GETTIME (time_end);
2066 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2068 /* FIXME: also report this to the user, preferably in gc-end. */
2069 if (major_collector.get_and_reset_num_major_objects_marked)
2070 major_collector.get_and_reset_num_major_objects_marked ();
2072 return bytes_pinned_from_failed_allocation > 0;
2076 major_start_concurrent_collection (const char *reason)
2078 TV_DECLARE (time_start);
2079 TV_DECLARE (time_end);
2080 long long num_objects_marked;
2082 if (disable_major_collections)
2085 TV_GETTIME (time_start);
2086 SGEN_TV_GETTIME (time_major_conc_collection_start);
2088 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2089 g_assert (num_objects_marked == 0);
2091 binary_protocol_concurrent_start ();
2093 // FIXME: store reason and pass it when finishing
2094 major_start_collection (TRUE, NULL);
2096 gray_queue_redirect (&gray_queue);
2098 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2100 TV_GETTIME (time_end);
2101 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2103 current_collection_generation = -1;
2107 * Returns whether the major collection has finished.
2110 major_should_finish_concurrent_collection (void)
2112 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2113 return sgen_workers_all_done ();
2117 major_update_concurrent_collection (void)
2119 TV_DECLARE (total_start);
2120 TV_DECLARE (total_end);
2122 TV_GETTIME (total_start);
2124 binary_protocol_concurrent_update ();
2126 major_collector.update_cardtable_mod_union ();
2127 sgen_los_update_cardtable_mod_union ();
2129 TV_GETTIME (total_end);
2130 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2134 major_finish_concurrent_collection (gboolean forced)
2136 TV_DECLARE (total_start);
2137 TV_DECLARE (total_end);
2139 TV_GETTIME (total_start);
2141 binary_protocol_concurrent_finish ();
2144 * We need to stop all workers since we're updating the cardtable below.
2145 * The workers will be resumed with a finishing pause context to avoid
2146 * additional cardtable and object scanning.
2148 sgen_workers_stop_all_workers ();
2150 SGEN_TV_GETTIME (time_major_conc_collection_end);
2151 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2153 major_collector.update_cardtable_mod_union ();
2154 sgen_los_update_cardtable_mod_union ();
2156 if (mod_union_consistency_check)
2157 sgen_check_mod_union_consistency ();
2159 current_collection_generation = GENERATION_OLD;
2160 sgen_cement_reset ();
2161 major_finish_collection ("finishing", -1, forced);
2163 if (whole_heap_check_before_collection)
2164 sgen_check_whole_heap (FALSE);
2166 TV_GETTIME (total_end);
2167 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2169 current_collection_generation = -1;
2173 * Ensure an allocation request for @size will succeed by freeing enough memory.
2175 * LOCKING: The GC lock MUST be held.
2178 sgen_ensure_free_space (size_t size, int generation)
2180 int generation_to_collect = -1;
2181 const char *reason = NULL;
2183 if (generation == GENERATION_OLD) {
2184 if (sgen_need_major_collection (size)) {
2185 reason = "LOS overflow";
2186 generation_to_collect = GENERATION_OLD;
2189 if (degraded_mode) {
2190 if (sgen_need_major_collection (size)) {
2191 reason = "Degraded mode overflow";
2192 generation_to_collect = GENERATION_OLD;
2194 } else if (sgen_need_major_collection (size)) {
2195 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2196 generation_to_collect = GENERATION_OLD;
2198 generation_to_collect = GENERATION_NURSERY;
2199 reason = "Nursery full";
2203 if (generation_to_collect == -1) {
2204 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2205 generation_to_collect = GENERATION_OLD;
2206 reason = "Finish concurrent collection";
2210 if (generation_to_collect == -1)
2212 sgen_perform_collection (size, generation_to_collect, reason, FALSE);
2216 * LOCKING: Assumes the GC lock is held.
2219 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish)
2221 TV_DECLARE (gc_start);
2222 TV_DECLARE (gc_end);
2223 TV_DECLARE (gc_total_start);
2224 TV_DECLARE (gc_total_end);
2225 GGTimingInfo infos [2];
2226 int overflow_generation_to_collect = -1;
2227 int oldest_generation_collected = generation_to_collect;
2228 const char *overflow_reason = NULL;
2230 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2232 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2234 TV_GETTIME (gc_start);
2236 sgen_stop_world (generation_to_collect);
2238 TV_GETTIME (gc_total_start);
2240 if (concurrent_collection_in_progress) {
2242 * If the concurrent worker is finished or we are asked to do a major collection
2243 * then we finish the concurrent collection.
2245 gboolean finish = major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD;
2248 major_finish_concurrent_collection (wait_to_finish);
2249 oldest_generation_collected = GENERATION_OLD;
2251 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY, "Why aren't we finishing the concurrent collection?");
2252 major_update_concurrent_collection ();
2253 collect_nursery (NULL, FALSE);
2259 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2262 * There's no concurrent collection in progress. Collect the generation we're asked
2263 * to collect. If the major collector is concurrent and we're not forced to wait,
2264 * start a concurrent collection.
2266 // FIXME: extract overflow reason
2267 if (generation_to_collect == GENERATION_NURSERY) {
2268 if (collect_nursery (NULL, FALSE)) {
2269 overflow_generation_to_collect = GENERATION_OLD;
2270 overflow_reason = "Minor overflow";
2273 if (major_collector.is_concurrent && !wait_to_finish) {
2274 collect_nursery (NULL, FALSE);
2275 major_start_concurrent_collection (reason);
2276 // FIXME: set infos[0] properly
2280 if (major_do_collection (reason, wait_to_finish)) {
2281 overflow_generation_to_collect = GENERATION_NURSERY;
2282 overflow_reason = "Excessive pinning";
2286 TV_GETTIME (gc_end);
2288 memset (infos, 0, sizeof (infos));
2289 infos [0].generation = generation_to_collect;
2290 infos [0].reason = reason;
2291 infos [0].is_overflow = FALSE;
2292 infos [1].generation = -1;
2293 infos [0].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2295 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2297 if (overflow_generation_to_collect != -1) {
2299 * We need to do an overflow collection, either because we ran out of memory
2300 * or the nursery is fully pinned.
2303 infos [1].generation = overflow_generation_to_collect;
2304 infos [1].reason = overflow_reason;
2305 infos [1].is_overflow = TRUE;
2308 if (overflow_generation_to_collect == GENERATION_NURSERY)
2309 collect_nursery (NULL, FALSE);
2311 major_do_collection (overflow_reason, wait_to_finish);
2313 TV_GETTIME (gc_end);
2314 infos [1].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2316 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2319 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2321 /* this also sets the proper pointers for the next allocation */
2322 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2323 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2324 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2325 sgen_dump_pin_queue ();
2330 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2332 TV_GETTIME (gc_total_end);
2333 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2335 sgen_restart_world (oldest_generation_collected, infos);
2339 * ######################################################################
2340 * ######## Memory allocation from the OS
2341 * ######################################################################
2342 * This section of code deals with getting memory from the OS and
2343 * allocating memory for GC-internal data structures.
2344 * Internal memory can be handled with a freelist for small objects.
2350 G_GNUC_UNUSED static void
2351 report_internal_mem_usage (void)
2353 printf ("Internal memory usage:\n");
2354 sgen_report_internal_mem_usage ();
2355 printf ("Pinned memory usage:\n");
2356 major_collector.report_pinned_memory_usage ();
2360 * ######################################################################
2361 * ######## Finalization support
2362 * ######################################################################
2366 * If the object has been forwarded it means it's still referenced from a root.
2367 * If it is pinned it's still alive as well.
2368 * A LOS object is only alive if we have pinned it.
2369 * Return TRUE if @obj is ready to be finalized.
2371 static inline gboolean
2372 sgen_is_object_alive (GCObject *object)
2374 if (ptr_in_nursery (object))
2375 return sgen_nursery_is_object_alive (object);
2377 return sgen_major_is_object_alive (object);
2381 * This function returns true if @object is either alive and belongs to the
2382 * current collection - major collections are full heap, so old gen objects
2383 * are never alive during a minor collection.
2386 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2388 if (ptr_in_nursery (object))
2389 return sgen_nursery_is_object_alive (object);
2391 if (current_collection_generation == GENERATION_NURSERY)
2394 return sgen_major_is_object_alive (object);
2399 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2401 return !sgen_is_object_alive (object);
2405 sgen_queue_finalization_entry (GCObject *obj)
2407 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2409 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2411 sgen_client_object_queued_for_finalization (obj);
2415 sgen_object_is_live (GCObject *obj)
2417 return sgen_is_object_alive_and_on_current_collection (obj);
2421 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2422 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2423 * all finalizers have really finished running.
2425 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2426 * This means that just checking whether the queues are empty leaves the possibility that an
2427 * object might have been dequeued but not yet finalized. That's why we need the additional
2428 * flag `pending_unqueued_finalizer`.
2431 static volatile gboolean pending_unqueued_finalizer = FALSE;
2434 sgen_gc_invoke_finalizers (void)
2438 g_assert (!pending_unqueued_finalizer);
2440 /* FIXME: batch to reduce lock contention */
2441 while (sgen_have_pending_finalizers ()) {
2447 * We need to set `pending_unqueued_finalizer` before dequeing the
2448 * finalizable object.
2450 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2451 pending_unqueued_finalizer = TRUE;
2452 mono_memory_write_barrier ();
2453 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2454 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2455 pending_unqueued_finalizer = TRUE;
2456 mono_memory_write_barrier ();
2457 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2463 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2471 /* the object is on the stack so it is pinned */
2472 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2473 sgen_client_run_finalize (obj);
2476 if (pending_unqueued_finalizer) {
2477 mono_memory_write_barrier ();
2478 pending_unqueued_finalizer = FALSE;
2485 sgen_have_pending_finalizers (void)
2487 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2491 * ######################################################################
2492 * ######## registered roots support
2493 * ######################################################################
2497 * We do not coalesce roots.
2500 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2502 RootRecord new_root;
2505 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2506 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2507 /* we allow changing the size and the descriptor (for thread statics etc) */
2509 size_t old_size = root->end_root - start;
2510 root->end_root = start + size;
2511 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2512 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2513 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2514 root->root_desc = descr;
2516 roots_size -= old_size;
2522 new_root.end_root = start + size;
2523 new_root.root_desc = descr;
2524 new_root.source = source;
2527 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2530 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);
2537 sgen_deregister_root (char* addr)
2543 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2544 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2545 roots_size -= (root.end_root - addr);
2551 * ######################################################################
2552 * ######## Thread handling (stop/start code)
2553 * ######################################################################
2557 sgen_get_current_collection_generation (void)
2559 return current_collection_generation;
2563 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2565 #ifndef HAVE_KW_THREAD
2566 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2569 sgen_init_tlab_info (info);
2571 sgen_client_thread_register (info, stack_bottom_fallback);
2577 sgen_thread_unregister (SgenThreadInfo *p)
2579 sgen_client_thread_unregister (p);
2583 * ######################################################################
2584 * ######## Write barriers
2585 * ######################################################################
2589 * Note: the write barriers first do the needed GC work and then do the actual store:
2590 * this way the value is visible to the conservative GC scan after the write barrier
2591 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2592 * the conservative scan, otherwise by the remembered set scan.
2596 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2598 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2599 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2600 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2601 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2605 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2606 if (binary_protocol_is_heavy_enabled ()) {
2608 for (i = 0; i < count; ++i) {
2609 gpointer dest = (gpointer*)dest_ptr + i;
2610 gpointer obj = *((gpointer*)src_ptr + i);
2612 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2617 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2621 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2625 HEAVY_STAT (++stat_wbarrier_generic_store);
2627 sgen_client_wbarrier_generic_nostore_check (ptr);
2629 obj = *(gpointer*)ptr;
2631 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2634 * We need to record old->old pointer locations for the
2635 * concurrent collector.
2637 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2638 SGEN_LOG (8, "Skipping remset at %p", ptr);
2642 SGEN_LOG (8, "Adding remset at %p", ptr);
2644 remset.wbarrier_generic_nostore (ptr);
2648 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2650 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2651 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2652 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2653 mono_gc_wbarrier_generic_nostore (ptr);
2654 sgen_dummy_use (value);
2657 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2658 * as an atomic operation with release semantics.
2661 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2663 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2665 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2667 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2669 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2670 mono_gc_wbarrier_generic_nostore (ptr);
2672 sgen_dummy_use (value);
2676 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2678 GCObject **dest = (GCObject **)_dest;
2679 GCObject **src = (GCObject **)_src;
2683 mono_gc_wbarrier_generic_store (dest, *src);
2688 size -= SIZEOF_VOID_P;
2694 * ######################################################################
2695 * ######## Other mono public interface functions.
2696 * ######################################################################
2700 sgen_gc_collect (int generation)
2705 sgen_perform_collection (0, generation, "user request", TRUE);
2710 sgen_gc_collection_count (int generation)
2712 if (generation == 0)
2713 return gc_stats.minor_gc_count;
2714 return gc_stats.major_gc_count;
2718 sgen_gc_get_used_size (void)
2722 tot = los_memory_usage;
2723 tot += nursery_section->next_data - nursery_section->data;
2724 tot += major_collector.get_used_size ();
2725 /* FIXME: account for pinned objects */
2731 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2735 va_start (ap, description_format);
2737 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2738 vfprintf (stderr, description_format, ap);
2740 fprintf (stderr, " - %s", fallback);
2741 fprintf (stderr, "\n");
2747 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2750 double val = strtod (opt, &endptr);
2751 if (endptr == opt) {
2752 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2755 else if (val < min || val > max) {
2756 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2768 char *major_collector_opt = NULL;
2769 char *minor_collector_opt = NULL;
2770 size_t max_heap = 0;
2771 size_t soft_limit = 0;
2773 gboolean debug_print_allowance = FALSE;
2774 double allowance_ratio = 0, save_target = 0;
2775 gboolean cement_enabled = TRUE;
2778 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2781 /* already inited */
2784 /* being inited by another thread */
2785 mono_thread_info_usleep (1000);
2788 /* we will init it */
2791 g_assert_not_reached ();
2793 } while (result != 0);
2795 SGEN_TV_GETTIME (sgen_init_timestamp);
2797 #ifdef SGEN_WITHOUT_MONO
2798 mono_thread_smr_init ();
2801 mono_coop_mutex_init (&gc_mutex);
2803 gc_debug_file = stderr;
2805 mono_coop_mutex_init (&sgen_interruption_mutex);
2807 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2808 opts = g_strsplit (env, ",", -1);
2809 for (ptr = opts; *ptr; ++ptr) {
2811 if (g_str_has_prefix (opt, "major=")) {
2812 opt = strchr (opt, '=') + 1;
2813 major_collector_opt = g_strdup (opt);
2814 } else if (g_str_has_prefix (opt, "minor=")) {
2815 opt = strchr (opt, '=') + 1;
2816 minor_collector_opt = g_strdup (opt);
2824 sgen_init_internal_allocator ();
2825 sgen_init_nursery_allocator ();
2826 sgen_init_fin_weak_hash ();
2827 sgen_init_hash_table ();
2828 sgen_init_descriptors ();
2829 sgen_init_gray_queues ();
2830 sgen_init_allocator ();
2831 sgen_init_gchandles ();
2833 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2834 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2836 sgen_client_init ();
2838 if (!minor_collector_opt) {
2839 sgen_simple_nursery_init (&sgen_minor_collector);
2841 if (!strcmp (minor_collector_opt, "simple")) {
2843 sgen_simple_nursery_init (&sgen_minor_collector);
2844 } else if (!strcmp (minor_collector_opt, "split")) {
2845 sgen_split_nursery_init (&sgen_minor_collector);
2847 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2848 goto use_simple_nursery;
2852 if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep")) {
2853 use_marksweep_major:
2854 sgen_marksweep_init (&major_collector);
2855 } else if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep-conc")) {
2856 sgen_marksweep_conc_init (&major_collector);
2858 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `marksweep` instead.", "Unknown major collector `%s'.", major_collector_opt);
2859 goto use_marksweep_major;
2862 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2865 gboolean usage_printed = FALSE;
2867 for (ptr = opts; *ptr; ++ptr) {
2869 if (!strcmp (opt, ""))
2871 if (g_str_has_prefix (opt, "major="))
2873 if (g_str_has_prefix (opt, "minor="))
2875 if (g_str_has_prefix (opt, "max-heap-size=")) {
2876 size_t page_size = mono_pagesize ();
2877 size_t max_heap_candidate = 0;
2878 opt = strchr (opt, '=') + 1;
2879 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2880 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2881 if (max_heap != max_heap_candidate)
2882 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2884 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2888 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2889 opt = strchr (opt, '=') + 1;
2890 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2891 if (soft_limit <= 0) {
2892 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2896 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2902 if (g_str_has_prefix (opt, "nursery-size=")) {
2904 opt = strchr (opt, '=') + 1;
2905 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2906 if ((val & (val - 1))) {
2907 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2911 if (val < SGEN_MAX_NURSERY_WASTE) {
2912 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2913 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2917 sgen_nursery_size = val;
2918 sgen_nursery_bits = 0;
2919 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2922 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2928 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2930 opt = strchr (opt, '=') + 1;
2931 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2932 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2937 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2939 opt = strchr (opt, '=') + 1;
2940 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2941 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2942 allowance_ratio = val;
2947 if (!strcmp (opt, "cementing")) {
2948 cement_enabled = TRUE;
2951 if (!strcmp (opt, "no-cementing")) {
2952 cement_enabled = FALSE;
2956 if (!strcmp (opt, "precleaning")) {
2957 precleaning_enabled = TRUE;
2960 if (!strcmp (opt, "no-precleaning")) {
2961 precleaning_enabled = FALSE;
2965 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2968 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2971 if (sgen_client_handle_gc_param (opt))
2974 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2979 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2980 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2981 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2982 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2983 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2984 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2985 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2986 fprintf (stderr, " [no-]cementing\n");
2987 if (major_collector.print_gc_param_usage)
2988 major_collector.print_gc_param_usage ();
2989 if (sgen_minor_collector.print_gc_param_usage)
2990 sgen_minor_collector.print_gc_param_usage ();
2991 sgen_client_print_gc_params_usage ();
2992 fprintf (stderr, " Experimental options:\n");
2993 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2994 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);
2995 fprintf (stderr, "\n");
2997 usage_printed = TRUE;
3002 if (major_collector_opt)
3003 g_free (major_collector_opt);
3005 if (minor_collector_opt)
3006 g_free (minor_collector_opt);
3010 sgen_cement_init (cement_enabled);
3012 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
3013 gboolean usage_printed = FALSE;
3015 opts = g_strsplit (env, ",", -1);
3016 for (ptr = opts; ptr && *ptr; ptr ++) {
3018 if (!strcmp (opt, ""))
3020 if (opt [0] >= '0' && opt [0] <= '9') {
3021 gc_debug_level = atoi (opt);
3026 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3027 gc_debug_file = fopen (rf, "wb");
3029 gc_debug_file = stderr;
3032 } else if (!strcmp (opt, "print-allowance")) {
3033 debug_print_allowance = TRUE;
3034 } else if (!strcmp (opt, "print-pinning")) {
3035 sgen_pin_stats_enable ();
3036 } else if (!strcmp (opt, "verify-before-allocs")) {
3037 verify_before_allocs = 1;
3038 has_per_allocation_action = TRUE;
3039 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3040 char *arg = strchr (opt, '=') + 1;
3041 verify_before_allocs = atoi (arg);
3042 has_per_allocation_action = TRUE;
3043 } else if (!strcmp (opt, "collect-before-allocs")) {
3044 collect_before_allocs = 1;
3045 has_per_allocation_action = TRUE;
3046 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3047 char *arg = strchr (opt, '=') + 1;
3048 has_per_allocation_action = TRUE;
3049 collect_before_allocs = atoi (arg);
3050 } else if (!strcmp (opt, "verify-before-collections")) {
3051 whole_heap_check_before_collection = TRUE;
3052 } else if (!strcmp (opt, "check-at-minor-collections")) {
3053 consistency_check_at_minor_collection = TRUE;
3054 nursery_clear_policy = CLEAR_AT_GC;
3055 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3056 if (!major_collector.is_concurrent) {
3057 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3060 mod_union_consistency_check = TRUE;
3061 } else if (!strcmp (opt, "check-mark-bits")) {
3062 check_mark_bits_after_major_collection = TRUE;
3063 } else if (!strcmp (opt, "check-nursery-pinned")) {
3064 check_nursery_objects_pinned = TRUE;
3065 } else if (!strcmp (opt, "clear-at-gc")) {
3066 nursery_clear_policy = CLEAR_AT_GC;
3067 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3068 nursery_clear_policy = CLEAR_AT_GC;
3069 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3070 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3071 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3072 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3073 } else if (!strcmp (opt, "check-scan-starts")) {
3074 do_scan_starts_check = TRUE;
3075 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3076 do_verify_nursery = TRUE;
3077 } else if (!strcmp (opt, "check-concurrent")) {
3078 if (!major_collector.is_concurrent) {
3079 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3082 nursery_clear_policy = CLEAR_AT_GC;
3083 do_concurrent_checks = TRUE;
3084 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3085 do_dump_nursery_content = TRUE;
3086 } else if (!strcmp (opt, "disable-minor")) {
3087 disable_minor_collections = TRUE;
3088 } else if (!strcmp (opt, "disable-major")) {
3089 disable_major_collections = TRUE;
3090 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3091 char *filename = strchr (opt, '=') + 1;
3092 nursery_clear_policy = CLEAR_AT_GC;
3093 sgen_debug_enable_heap_dump (filename);
3094 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3095 char *filename = strchr (opt, '=') + 1;
3096 char *colon = strrchr (filename, ':');
3099 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3100 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3105 binary_protocol_init (filename, (long long)limit);
3106 } else if (!strcmp (opt, "nursery-canaries")) {
3107 do_verify_nursery = TRUE;
3108 enable_nursery_canaries = TRUE;
3109 } else if (!sgen_client_handle_gc_debug (opt)) {
3110 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3115 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);
3116 fprintf (stderr, "Valid <option>s are:\n");
3117 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3118 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3119 fprintf (stderr, " check-at-minor-collections\n");
3120 fprintf (stderr, " check-mark-bits\n");
3121 fprintf (stderr, " check-nursery-pinned\n");
3122 fprintf (stderr, " verify-before-collections\n");
3123 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3124 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3125 fprintf (stderr, " disable-minor\n");
3126 fprintf (stderr, " disable-major\n");
3127 fprintf (stderr, " check-concurrent\n");
3128 fprintf (stderr, " clear-[nursery-]at-gc\n");
3129 fprintf (stderr, " clear-at-tlab-creation\n");
3130 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3131 fprintf (stderr, " check-scan-starts\n");
3132 fprintf (stderr, " print-allowance\n");
3133 fprintf (stderr, " print-pinning\n");
3134 fprintf (stderr, " heap-dump=<filename>\n");
3135 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3136 fprintf (stderr, " nursery-canaries\n");
3137 sgen_client_print_gc_debug_usage ();
3138 fprintf (stderr, "\n");
3140 usage_printed = TRUE;
3146 if (check_mark_bits_after_major_collection)
3147 nursery_clear_policy = CLEAR_AT_GC;
3149 if (major_collector.post_param_init)
3150 major_collector.post_param_init (&major_collector);
3152 if (major_collector.needs_thread_pool)
3153 sgen_workers_init (1);
3155 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3157 memset (&remset, 0, sizeof (remset));
3159 sgen_card_table_init (&remset);
3161 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");
3167 sgen_get_nursery_clear_policy (void)
3169 return nursery_clear_policy;
3175 mono_coop_mutex_lock (&gc_mutex);
3179 sgen_gc_unlock (void)
3181 mono_coop_mutex_unlock (&gc_mutex);
3185 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3187 major_collector.iterate_live_block_ranges (callback);
3191 sgen_get_major_collector (void)
3193 return &major_collector;
3197 sgen_get_remset (void)
3203 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3205 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3206 sgen_los_count_cards (los_total, los_marked);
3209 static gboolean world_is_stopped = FALSE;
3211 /* LOCKING: assumes the GC lock is held */
3213 sgen_stop_world (int generation)
3215 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3217 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3219 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3221 sgen_client_stop_world (generation);
3223 world_is_stopped = TRUE;
3225 if (binary_protocol_is_heavy_enabled ())
3226 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3227 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3230 /* LOCKING: assumes the GC lock is held */
3232 sgen_restart_world (int generation, GGTimingInfo *timing)
3234 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3236 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3238 if (binary_protocol_is_heavy_enabled ())
3239 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3240 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3242 sgen_client_restart_world (generation, timing);
3244 world_is_stopped = FALSE;
3246 binary_protocol_world_restarted (generation, sgen_timestamp ());
3248 if (sgen_client_bridge_need_processing ())
3249 sgen_client_bridge_processing_finish (generation);
3251 sgen_memgov_collection_end (generation, timing, timing ? 2 : 0);
3255 sgen_is_world_stopped (void)
3257 return world_is_stopped;
3261 sgen_check_whole_heap_stw (void)
3263 sgen_stop_world (0);
3264 sgen_clear_nursery_fragments ();
3265 sgen_check_whole_heap (FALSE);
3266 sgen_restart_world (0, NULL);
3270 sgen_timestamp (void)
3272 SGEN_TV_DECLARE (timestamp);
3273 SGEN_TV_GETTIME (timestamp);
3274 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3277 #endif /* HAVE_SGEN_GC */