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;
346 gboolean sgen_try_free_some_memory;
348 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
350 size_t degraded_mode = 0;
352 static mword bytes_pinned_from_failed_allocation = 0;
354 GCMemSection *nursery_section = NULL;
355 static volatile mword lowest_heap_address = ~(mword)0;
356 static volatile mword highest_heap_address = 0;
358 MonoCoopMutex sgen_interruption_mutex;
360 int current_collection_generation = -1;
361 static volatile gboolean concurrent_collection_in_progress = FALSE;
363 /* objects that are ready to be finalized */
364 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
365 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
367 /* registered roots: the key to the hash is the root start address */
369 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
371 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
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),
374 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
376 static mword roots_size = 0; /* amount of memory in the root set */
378 /* The size of a TLAB */
379 /* The bigger the value, the less often we have to go to the slow path to allocate a new
380 * one, but the more space is wasted by threads not allocating much memory.
382 * FIXME: Make this self-tuning for each thread.
384 guint32 tlab_size = (1024 * 4);
386 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
388 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
390 #define ALIGN_UP SGEN_ALIGN_UP
392 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
393 MonoNativeThreadId main_gc_thread = NULL;
396 /*Object was pinned during the current collection*/
397 static mword objects_pinned;
400 * ######################################################################
401 * ######## Macros and function declarations.
402 * ######################################################################
405 typedef SgenGrayQueue GrayQueue;
407 /* forward declarations */
408 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
410 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
411 static void finish_gray_stack (int generation, ScanCopyContext ctx);
414 SgenMajorCollector major_collector;
415 SgenMinorCollector sgen_minor_collector;
416 /* FIXME: get rid of this */
417 static GrayQueue gray_queue;
419 static SgenRememberedSet remset;
421 /* The gray queue to use from the main collection thread. */
422 #define WORKERS_DISTRIBUTE_GRAY_QUEUE (&gray_queue)
425 * The gray queue a worker job must use. If we're not parallel or
426 * concurrent, we use the main gray queue.
428 static SgenGrayQueue*
429 sgen_workers_get_job_gray_queue (WorkerData *worker_data)
431 return worker_data ? &worker_data->private_gray_queue : WORKERS_DISTRIBUTE_GRAY_QUEUE;
435 gray_queue_redirect (SgenGrayQueue *queue)
437 gboolean wake = FALSE;
440 GrayQueueSection *section = sgen_gray_object_dequeue_section (queue);
443 sgen_section_gray_queue_enqueue ((SgenSectionGrayQueue *)queue->alloc_prepare_data, section);
448 g_assert (concurrent_collection_in_progress);
449 sgen_workers_ensure_awake ();
454 gray_queue_enable_redirect (SgenGrayQueue *queue)
456 if (!concurrent_collection_in_progress)
459 sgen_gray_queue_set_alloc_prepare (queue, gray_queue_redirect, sgen_workers_get_distribute_section_gray_queue ());
460 gray_queue_redirect (queue);
464 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
466 while (start < end) {
470 if (!*(void**)start) {
471 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
476 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
482 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
483 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
484 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
485 callback ((GCObject*)obj, size, data);
486 CANARIFY_SIZE (size);
488 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
496 * sgen_add_to_global_remset:
498 * The global remset contains locations which point into newspace after
499 * a minor collection. This can happen if the objects they point to are pinned.
501 * LOCKING: If called from a parallel collector, the global remset
502 * lock must be held. For serial collectors that is not necessary.
505 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
507 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
509 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
511 if (!major_collector.is_concurrent) {
512 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
514 if (current_collection_generation == -1)
515 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
518 if (!object_is_pinned (obj))
519 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");
520 else if (sgen_cement_lookup_or_register (obj))
523 remset.record_pointer (ptr);
525 sgen_pin_stats_register_global_remset (obj);
527 SGEN_LOG (8, "Adding global remset for %p", ptr);
528 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
532 * sgen_drain_gray_stack:
534 * Scan objects in the gray stack until the stack is empty. This should be called
535 * frequently after each object is copied, to achieve better locality and cache
540 sgen_drain_gray_stack (ScanCopyContext ctx)
542 ScanObjectFunc scan_func = ctx.ops->scan_object;
543 GrayQueue *queue = ctx.queue;
545 if (ctx.ops->drain_gray_stack)
546 return ctx.ops->drain_gray_stack (queue);
551 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
554 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
555 scan_func (obj, desc, queue);
561 * Addresses in the pin queue are already sorted. This function finds
562 * the object header for each address and pins the object. The
563 * addresses must be inside the nursery section. The (start of the)
564 * address array is overwritten with the addresses of the actually
565 * pinned objects. Return the number of pinned objects.
568 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
570 GCMemSection *section = nursery_section;
571 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
572 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
573 void *start_nursery = section->data;
574 void *end_nursery = section->next_data;
579 void *pinning_front = start_nursery;
581 void **definitely_pinned = start;
582 ScanObjectFunc scan_func = ctx.ops->scan_object;
583 SgenGrayQueue *queue = ctx.queue;
585 sgen_nursery_allocator_prepare_for_pinning ();
587 while (start < end) {
588 GCObject *obj_to_pin = NULL;
589 size_t obj_to_pin_size = 0;
594 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
595 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
602 SGEN_LOG (5, "Considering pinning addr %p", addr);
603 /* We've already processed everything up to pinning_front. */
604 if (addr < pinning_front) {
610 * Find the closest scan start <= addr. We might search backward in the
611 * scan_starts array because entries might be NULL. In the worst case we
612 * start at start_nursery.
614 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
615 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
616 search_start = (void*)section->scan_starts [idx];
617 if (!search_start || search_start > addr) {
620 search_start = section->scan_starts [idx];
621 if (search_start && search_start <= addr)
624 if (!search_start || search_start > addr)
625 search_start = start_nursery;
629 * If the pinning front is closer than the scan start we found, start
630 * searching at the front.
632 if (search_start < pinning_front)
633 search_start = pinning_front;
636 * Now addr should be in an object a short distance from search_start.
638 * search_start must point to zeroed mem or point to an object.
641 size_t obj_size, canarified_obj_size;
644 if (!*(void**)search_start) {
645 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
646 /* The loop condition makes sure we don't overrun addr. */
650 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
653 * Filler arrays are marked by an invalid sync word. We don't
654 * consider them for pinning. They are not delimited by canaries,
657 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
658 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
659 CANARIFY_SIZE (canarified_obj_size);
661 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
662 /* This is the object we're looking for. */
663 obj_to_pin = (GCObject*)search_start;
664 obj_to_pin_size = canarified_obj_size;
669 /* Skip to the next object */
670 search_start = (void*)((char*)search_start + canarified_obj_size);
671 } while (search_start <= addr);
673 /* We've searched past the address we were looking for. */
675 pinning_front = search_start;
676 goto next_pin_queue_entry;
680 * We've found an object to pin. It might still be a dummy array, but we
681 * can advance the pinning front in any case.
683 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
686 * If this is a dummy array marking the beginning of a nursery
687 * fragment, we don't pin it.
689 if (sgen_client_object_is_array_fill (obj_to_pin))
690 goto next_pin_queue_entry;
693 * Finally - pin the object!
695 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
696 if (do_scan_objects) {
697 scan_func (obj_to_pin, desc, queue);
699 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
700 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
701 binary_protocol_pin (obj_to_pin,
702 (gpointer)LOAD_VTABLE (obj_to_pin),
703 safe_object_get_size (obj_to_pin));
705 pin_object (obj_to_pin);
706 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
707 sgen_pin_stats_register_object (obj_to_pin, obj_to_pin_size);
708 definitely_pinned [count] = obj_to_pin;
711 if (concurrent_collection_in_progress)
712 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
714 next_pin_queue_entry:
718 sgen_client_nursery_objects_pinned (definitely_pinned, count);
719 stat_pinned_objects += count;
724 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
728 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
731 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
732 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
736 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
737 * when we can't promote an object because we're out of memory.
740 sgen_pin_object (GCObject *object, GrayQueue *queue)
743 * All pinned objects are assumed to have been staged, so we need to stage as well.
744 * Also, the count of staged objects shows that "late pinning" happened.
746 sgen_pin_stage_ptr (object);
748 SGEN_PIN_OBJECT (object);
749 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
752 sgen_pin_stats_register_object (object, safe_object_get_size (object));
754 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
757 /* Sort the addresses in array in increasing order.
758 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
761 sgen_sort_addresses (void **array, size_t size)
766 for (i = 1; i < size; ++i) {
769 size_t parent = (child - 1) / 2;
771 if (array [parent] >= array [child])
774 tmp = array [parent];
775 array [parent] = array [child];
782 for (i = size - 1; i > 0; --i) {
785 array [i] = array [0];
791 while (root * 2 + 1 <= end) {
792 size_t child = root * 2 + 1;
794 if (child < end && array [child] < array [child + 1])
796 if (array [root] >= array [child])
800 array [root] = array [child];
809 * Scan the memory between start and end and queue values which could be pointers
810 * to the area between start_nursery and end_nursery for later consideration.
811 * Typically used for thread stacks.
814 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
818 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
820 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
821 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
824 while (start < end) {
826 * *start can point to the middle of an object
827 * note: should we handle pointing at the end of an object?
828 * pinning in C# code disallows pointing at the end of an object
829 * but there is some small chance that an optimizing C compiler
830 * may keep the only reference to an object by pointing
831 * at the end of it. We ignore this small chance for now.
832 * Pointers to the end of an object are indistinguishable
833 * from pointers to the start of the next object in memory
834 * so if we allow that we'd need to pin two objects...
835 * We queue the pointer in an array, the
836 * array will then be sorted and uniqued. This way
837 * we can coalesce several pinning pointers and it should
838 * be faster since we'd do a memory scan with increasing
839 * addresses. Note: we can align the address to the allocation
840 * alignment, so the unique process is more effective.
842 mword addr = (mword)*start;
843 addr &= ~(ALLOC_ALIGN - 1);
844 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
845 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
846 sgen_pin_stage_ptr ((void*)addr);
847 binary_protocol_pin_stage (start, (void*)addr);
848 sgen_pin_stats_register_address ((char*)addr, pin_type);
854 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
858 * The first thing we do in a collection is to identify pinned objects.
859 * This function considers all the areas of memory that need to be
860 * conservatively scanned.
863 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
867 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);
868 /* objects pinned from the API are inside these roots */
869 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
870 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
871 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
872 } SGEN_HASH_TABLE_FOREACH_END;
873 /* now deal with the thread stacks
874 * in the future we should be able to conservatively scan only:
875 * *) the cpu registers
876 * *) the unmanaged stack frames
877 * *) the _last_ managed stack frame
878 * *) pointers slots in managed frames
880 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
884 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
886 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
887 ctx->ops->copy_or_mark_object (obj, ctx->queue);
891 * The memory area from start_root to end_root contains pointers to objects.
892 * Their position is precisely described by @desc (this means that the pointer
893 * can be either NULL or the pointer to the start of an object).
894 * This functions copies them to to_space updates them.
896 * This function is not thread-safe!
899 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
901 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
902 SgenGrayQueue *queue = ctx.queue;
904 switch (desc & ROOT_DESC_TYPE_MASK) {
905 case ROOT_DESC_BITMAP:
906 desc >>= ROOT_DESC_TYPE_SHIFT;
908 if ((desc & 1) && *start_root) {
909 copy_func ((GCObject**)start_root, queue);
910 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
916 case ROOT_DESC_COMPLEX: {
917 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
918 gsize bwords = (*bitmap_data) - 1;
919 void **start_run = start_root;
921 while (bwords-- > 0) {
922 gsize bmap = *bitmap_data++;
923 void **objptr = start_run;
925 if ((bmap & 1) && *objptr) {
926 copy_func ((GCObject**)objptr, queue);
927 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
932 start_run += GC_BITS_PER_WORD;
936 case ROOT_DESC_USER: {
937 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
938 marker (start_root, single_arg_user_copy_or_mark, &ctx);
941 case ROOT_DESC_RUN_LEN:
942 g_assert_not_reached ();
944 g_assert_not_reached ();
949 reset_heap_boundaries (void)
951 lowest_heap_address = ~(mword)0;
952 highest_heap_address = 0;
956 sgen_update_heap_boundaries (mword low, mword high)
961 old = lowest_heap_address;
964 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
967 old = highest_heap_address;
970 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
974 * Allocate and setup the data structures needed to be able to allocate objects
975 * in the nursery. The nursery is stored in nursery_section.
980 GCMemSection *section;
987 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
988 /* later we will alloc a larger area for the nursery but only activate
989 * what we need. The rest will be used as expansion if we have too many pinned
990 * objects in the existing nursery.
992 /* FIXME: handle OOM */
993 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
995 alloc_size = sgen_nursery_size;
997 /* If there isn't enough space even for the nursery we should simply abort. */
998 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
1000 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
1001 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
1002 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 ());
1003 section->data = section->next_data = data;
1004 section->size = alloc_size;
1005 section->end_data = data + sgen_nursery_size;
1006 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
1007 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
1008 section->num_scan_start = scan_starts;
1010 nursery_section = section;
1012 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
1016 mono_gc_get_logfile (void)
1018 return gc_debug_file;
1022 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1024 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1025 SgenGrayQueue *queue = ctx.queue;
1028 for (i = 0; i < fin_queue->next_slot; ++i) {
1029 GCObject *obj = (GCObject *)fin_queue->data [i];
1032 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1033 copy_func ((GCObject**)&fin_queue->data [i], queue);
1038 generation_name (int generation)
1040 switch (generation) {
1041 case GENERATION_NURSERY: return "nursery";
1042 case GENERATION_OLD: return "old";
1043 default: g_assert_not_reached ();
1048 sgen_generation_name (int generation)
1050 return generation_name (generation);
1054 finish_gray_stack (int generation, ScanCopyContext ctx)
1058 int done_with_ephemerons, ephemeron_rounds = 0;
1059 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1060 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1061 SgenGrayQueue *queue = ctx.queue;
1063 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1065 * We copied all the reachable objects. Now it's the time to copy
1066 * the objects that were not referenced by the roots, but by the copied objects.
1067 * we built a stack of objects pointed to by gray_start: they are
1068 * additional roots and we may add more items as we go.
1069 * We loop until gray_start == gray_objects which means no more objects have
1070 * been added. Note this is iterative: no recursion is involved.
1071 * We need to walk the LO list as well in search of marked big objects
1072 * (use a flag since this is needed only on major collections). We need to loop
1073 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1074 * To achieve better cache locality and cache usage, we drain the gray stack
1075 * frequently, after each object is copied, and just finish the work here.
1077 sgen_drain_gray_stack (ctx);
1079 SGEN_LOG (2, "%s generation done", generation_name (generation));
1082 Reset bridge data, we might have lingering data from a previous collection if this is a major
1083 collection trigged by minor overflow.
1085 We must reset the gathered bridges since their original block might be evacuated due to major
1086 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1088 if (sgen_client_bridge_need_processing ())
1089 sgen_client_bridge_reset_data ();
1092 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1093 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1094 * objects that are in fact reachable.
1096 done_with_ephemerons = 0;
1098 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1099 sgen_drain_gray_stack (ctx);
1101 } while (!done_with_ephemerons);
1103 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1105 if (sgen_client_bridge_need_processing ()) {
1106 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1107 sgen_drain_gray_stack (ctx);
1108 sgen_collect_bridge_objects (generation, ctx);
1109 if (generation == GENERATION_OLD)
1110 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1113 Do the first bridge step here, as the collector liveness state will become useless after that.
1115 An important optimization is to only proccess the possibly dead part of the object graph and skip
1116 over all live objects as we transitively know everything they point must be alive too.
1118 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1120 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1121 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1124 sgen_client_bridge_processing_stw_step ();
1128 Make sure we drain the gray stack before processing disappearing links and finalizers.
1129 If we don't make sure it is empty we might wrongly see a live object as dead.
1131 sgen_drain_gray_stack (ctx);
1134 We must clear weak links that don't track resurrection before processing object ready for
1135 finalization so they can be cleared before that.
1137 sgen_null_link_in_range (generation, ctx, FALSE);
1138 if (generation == GENERATION_OLD)
1139 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1142 /* walk the finalization queue and move also the objects that need to be
1143 * finalized: use the finalized objects as new roots so the objects they depend
1144 * on are also not reclaimed. As with the roots above, only objects in the nursery
1145 * are marked/copied.
1147 sgen_finalize_in_range (generation, ctx);
1148 if (generation == GENERATION_OLD)
1149 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1150 /* drain the new stack that might have been created */
1151 SGEN_LOG (6, "Precise scan of gray area post fin");
1152 sgen_drain_gray_stack (ctx);
1155 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1157 done_with_ephemerons = 0;
1159 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1160 sgen_drain_gray_stack (ctx);
1162 } while (!done_with_ephemerons);
1164 sgen_client_clear_unreachable_ephemerons (ctx);
1167 * We clear togglerefs only after all possible chances of revival are done.
1168 * This is semantically more inline with what users expect and it allows for
1169 * user finalizers to correctly interact with TR objects.
1171 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1174 SGEN_LOG (2, "Finalize queue handling scan for %s generation: %lld usecs %d ephemeron rounds", generation_name (generation), TV_ELAPSED (atv, btv), ephemeron_rounds);
1177 * handle disappearing links
1178 * Note we do this after checking the finalization queue because if an object
1179 * survives (at least long enough to be finalized) we don't clear the link.
1180 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1181 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1184 g_assert (sgen_gray_object_queue_is_empty (queue));
1186 sgen_null_link_in_range (generation, ctx, TRUE);
1187 if (generation == GENERATION_OLD)
1188 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1189 if (sgen_gray_object_queue_is_empty (queue))
1191 sgen_drain_gray_stack (ctx);
1194 g_assert (sgen_gray_object_queue_is_empty (queue));
1196 sgen_gray_object_queue_trim_free_list (queue);
1197 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1201 sgen_check_section_scan_starts (GCMemSection *section)
1204 for (i = 0; i < section->num_scan_start; ++i) {
1205 if (section->scan_starts [i]) {
1206 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1207 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1213 check_scan_starts (void)
1215 if (!do_scan_starts_check)
1217 sgen_check_section_scan_starts (nursery_section);
1218 major_collector.check_scan_starts ();
1222 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1226 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1227 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1228 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1229 } SGEN_HASH_TABLE_FOREACH_END;
1235 static gboolean inited = FALSE;
1240 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1242 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1243 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1244 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1245 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1246 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1247 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1249 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1250 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1251 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1252 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1253 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1254 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1255 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1256 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1257 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1258 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1260 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1262 #ifdef HEAVY_STATISTICS
1263 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1264 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1265 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1266 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1267 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1269 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1270 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1272 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1273 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1274 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1275 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1277 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1278 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1280 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1282 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1283 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1284 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1285 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1287 sgen_nursery_allocator_init_heavy_stats ();
1295 reset_pinned_from_failed_allocation (void)
1297 bytes_pinned_from_failed_allocation = 0;
1301 sgen_set_pinned_from_failed_allocation (mword objsize)
1303 bytes_pinned_from_failed_allocation += objsize;
1307 sgen_collection_is_concurrent (void)
1309 switch (current_collection_generation) {
1310 case GENERATION_NURSERY:
1312 case GENERATION_OLD:
1313 return concurrent_collection_in_progress;
1315 g_error ("Invalid current generation %d", current_collection_generation);
1321 sgen_concurrent_collection_in_progress (void)
1323 return concurrent_collection_in_progress;
1327 SgenThreadPoolJob job;
1328 SgenObjectOperations *ops;
1332 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1334 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1335 ScanJob *job_data = (ScanJob*)job;
1336 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1337 remset.scan_remsets (ctx);
1341 SgenThreadPoolJob job;
1342 SgenObjectOperations *ops;
1346 } ScanFromRegisteredRootsJob;
1349 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1351 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1352 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1353 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1355 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1359 SgenThreadPoolJob job;
1360 SgenObjectOperations *ops;
1363 } ScanThreadDataJob;
1366 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1368 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1369 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1370 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1372 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1376 SgenThreadPoolJob job;
1377 SgenObjectOperations *ops;
1378 SgenPointerQueue *queue;
1379 } ScanFinalizerEntriesJob;
1382 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1384 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1385 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1386 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1388 scan_finalizer_entries (job_data->queue, ctx);
1392 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1394 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1395 ScanJob *job_data = (ScanJob*)job;
1396 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1398 g_assert (concurrent_collection_in_progress);
1399 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1403 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1405 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1406 ScanJob *job_data = (ScanJob*)job;
1407 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1409 g_assert (concurrent_collection_in_progress);
1410 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1414 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1416 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1417 ScanJob *job_data = (ScanJob*)job;
1418 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1420 g_assert (concurrent_collection_in_progress);
1422 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1423 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1425 sgen_scan_pin_queue_objects (ctx);
1429 init_gray_queue (gboolean use_workers)
1432 sgen_workers_init_distribute_gray_queue ();
1433 sgen_gray_object_queue_init (&gray_queue, NULL);
1437 enqueue_scan_from_roots_jobs (char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1439 ScanFromRegisteredRootsJob *scrrj;
1440 ScanThreadDataJob *stdj;
1441 ScanFinalizerEntriesJob *sfej;
1443 /* registered roots, this includes static fields */
1445 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1447 scrrj->heap_start = heap_start;
1448 scrrj->heap_end = heap_end;
1449 scrrj->root_type = ROOT_TYPE_NORMAL;
1450 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1452 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1454 scrrj->heap_start = heap_start;
1455 scrrj->heap_end = heap_end;
1456 scrrj->root_type = ROOT_TYPE_WBARRIER;
1457 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1461 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1462 stdj->heap_start = heap_start;
1463 stdj->heap_end = heap_end;
1464 sgen_workers_enqueue_job (&stdj->job, enqueue);
1466 /* Scan the list of objects ready for finalization. */
1468 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1469 sfej->queue = &fin_ready_queue;
1471 sgen_workers_enqueue_job (&sfej->job, enqueue);
1473 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1474 sfej->queue = &critical_fin_queue;
1476 sgen_workers_enqueue_job (&sfej->job, enqueue);
1480 * Perform a nursery collection.
1482 * Return whether any objects were late-pinned due to being out of memory.
1485 collect_nursery (SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1487 gboolean needs_major;
1488 size_t max_garbage_amount;
1490 mword fragment_total;
1492 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1493 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue);
1497 if (disable_minor_collections)
1500 TV_GETTIME (last_minor_collection_start_tv);
1501 atv = last_minor_collection_start_tv;
1503 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1505 if (do_verify_nursery || do_dump_nursery_content)
1506 sgen_debug_verify_nursery (do_dump_nursery_content);
1508 current_collection_generation = GENERATION_NURSERY;
1510 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1512 reset_pinned_from_failed_allocation ();
1514 check_scan_starts ();
1516 sgen_nursery_alloc_prepare_for_minor ();
1520 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1521 /* FIXME: optimize later to use the higher address where an object can be present */
1522 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1524 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 ()));
1525 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1526 g_assert (nursery_section->size >= max_garbage_amount);
1528 /* world must be stopped already */
1530 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1532 sgen_client_pre_collection_checks ();
1534 nursery_section->next_data = nursery_next;
1536 major_collector.start_nursery_collection ();
1538 sgen_memgov_minor_collection_start ();
1540 init_gray_queue (FALSE);
1542 gc_stats.minor_gc_count ++;
1544 if (whole_heap_check_before_collection) {
1545 sgen_clear_nursery_fragments ();
1546 sgen_check_whole_heap (finish_up_concurrent_mark);
1548 if (consistency_check_at_minor_collection)
1549 sgen_check_consistency ();
1551 sgen_process_fin_stage_entries ();
1553 /* pin from pinned handles */
1554 sgen_init_pinning ();
1555 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1556 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1557 /* pin cemented objects */
1558 sgen_pin_cemented_objects ();
1559 /* identify pinned objects */
1560 sgen_optimize_pin_queue ();
1561 sgen_pinning_setup_section (nursery_section);
1563 pin_objects_in_nursery (FALSE, ctx);
1564 sgen_pinning_trim_queue_to_section (nursery_section);
1567 time_minor_pinning += TV_ELAPSED (btv, atv);
1568 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), TV_ELAPSED (btv, atv));
1569 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1571 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1572 sj->ops = object_ops;
1573 sgen_workers_enqueue_job (&sj->job, FALSE);
1575 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1577 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1578 SGEN_LOG (2, "Old generation scan: %lld usecs", TV_ELAPSED (atv, btv));
1580 sgen_pin_stats_print_class_stats ();
1582 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1583 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1586 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1588 enqueue_scan_from_roots_jobs (sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1591 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1593 finish_gray_stack (GENERATION_NURSERY, ctx);
1596 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1597 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1599 if (objects_pinned) {
1600 sgen_optimize_pin_queue ();
1601 sgen_pinning_setup_section (nursery_section);
1604 /* walk the pin_queue, build up the fragment list of free memory, unmark
1605 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1608 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1609 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1610 if (!fragment_total)
1613 /* Clear TLABs for all threads */
1614 sgen_clear_tlabs ();
1616 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1618 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1619 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1621 if (consistency_check_at_minor_collection)
1622 sgen_check_major_refs ();
1624 major_collector.finish_nursery_collection ();
1626 TV_GETTIME (last_minor_collection_end_tv);
1627 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1629 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1631 /* prepare the pin queue for the next collection */
1632 sgen_finish_pinning ();
1633 if (sgen_have_pending_finalizers ()) {
1634 SGEN_LOG (4, "Finalizer-thread wakeup");
1635 sgen_client_finalize_notify ();
1637 sgen_pin_stats_reset ();
1638 /* clear cemented hash */
1639 sgen_cement_clear_below_threshold ();
1641 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
1643 remset.finish_minor_collection ();
1645 check_scan_starts ();
1647 binary_protocol_flush_buffers (FALSE);
1649 sgen_memgov_minor_collection_end ();
1651 /*objects are late pinned because of lack of memory, so a major is a good call*/
1652 needs_major = objects_pinned > 0;
1653 current_collection_generation = -1;
1656 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1658 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1659 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1665 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1666 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1667 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1668 } CopyOrMarkFromRootsMode;
1671 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1676 /* FIXME: only use these values for the precise scan
1677 * note that to_space pointers should be excluded anyway...
1679 char *heap_start = NULL;
1680 char *heap_end = (char*)-1;
1681 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, WORKERS_DISTRIBUTE_GRAY_QUEUE);
1682 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1684 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1686 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1687 /*This cleans up unused fragments */
1688 sgen_nursery_allocator_prepare_for_pinning ();
1690 if (do_concurrent_checks)
1691 sgen_debug_check_nursery_is_clean ();
1693 /* The concurrent collector doesn't touch the nursery. */
1694 sgen_nursery_alloc_prepare_for_major ();
1697 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1701 /* Pinning depends on this */
1702 sgen_clear_nursery_fragments ();
1704 if (whole_heap_check_before_collection)
1705 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1708 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1710 if (!sgen_collection_is_concurrent ())
1711 nursery_section->next_data = sgen_get_nursery_end ();
1712 /* we should also coalesce scanning from sections close to each other
1713 * and deal with pointers outside of the sections later.
1718 sgen_client_pre_collection_checks ();
1721 /* Remsets are not useful for a major collection */
1722 remset.clear_cards ();
1725 sgen_process_fin_stage_entries ();
1728 sgen_init_pinning ();
1729 SGEN_LOG (6, "Collecting pinned addresses");
1730 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1731 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1732 /* Pin cemented objects that were forced */
1733 sgen_pin_cemented_objects ();
1735 sgen_optimize_pin_queue ();
1736 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1738 * Cemented objects that are in the pinned list will be marked. When
1739 * marking concurrently we won't mark mod-union cards for these objects.
1740 * Instead they will remain cemented until the next major collection,
1741 * when we will recheck if they are still pinned in the roots.
1743 sgen_cement_force_pinned ();
1746 sgen_client_collecting_major_1 ();
1749 * pin_queue now contains all candidate pointers, sorted and
1750 * uniqued. We must do two passes now to figure out which
1751 * objects are pinned.
1753 * The first is to find within the pin_queue the area for each
1754 * section. This requires that the pin_queue be sorted. We
1755 * also process the LOS objects and pinned chunks here.
1757 * The second, destructive, pass is to reduce the section
1758 * areas to pointers to the actually pinned objects.
1760 SGEN_LOG (6, "Pinning from sections");
1761 /* first pass for the sections */
1762 sgen_find_section_pin_queue_start_end (nursery_section);
1763 /* identify possible pointers to the insize of large objects */
1764 SGEN_LOG (6, "Pinning from large objects");
1765 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1767 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1768 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1770 if (sgen_los_object_is_pinned (bigobj->data)) {
1771 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1774 sgen_los_pin_object (bigobj->data);
1775 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1776 GRAY_OBJECT_ENQUEUE (WORKERS_DISTRIBUTE_GRAY_QUEUE, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1777 sgen_pin_stats_register_object (bigobj->data, safe_object_get_size (bigobj->data));
1778 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1779 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1780 (unsigned long)sgen_los_object_size (bigobj));
1782 sgen_client_pinned_los_object (bigobj->data);
1786 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1787 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1788 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1790 major_collector.pin_objects (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1791 if (old_next_pin_slot)
1792 *old_next_pin_slot = sgen_get_pinned_count ();
1795 time_major_pinning += TV_ELAPSED (atv, btv);
1796 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), TV_ELAPSED (atv, btv));
1797 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1799 major_collector.init_to_space ();
1801 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1802 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1803 if (sgen_workers_have_idle_work ()) {
1805 * We force the finish of the worker with the new object ops context
1806 * which can also do copying. We need to have finished pinning.
1808 sgen_workers_start_all_workers (object_ops, NULL);
1809 sgen_workers_join ();
1813 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1814 main_gc_thread = mono_native_thread_self ();
1817 sgen_client_collecting_major_2 ();
1820 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1822 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1824 enqueue_scan_from_roots_jobs (heap_start, heap_end, object_ops, FALSE);
1827 time_major_scan_roots += TV_ELAPSED (atv, btv);
1830 * We start the concurrent worker after pinning and after we scanned the roots
1831 * in order to make sure that the worker does not finish before handling all
1834 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1835 if (precleaning_enabled) {
1837 /* Mod union preclean job */
1838 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1839 sj->ops = object_ops;
1840 sgen_workers_start_all_workers (object_ops, &sj->job);
1842 sgen_workers_start_all_workers (object_ops, NULL);
1844 gray_queue_enable_redirect (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1847 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1850 /* Mod union card table */
1851 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1852 sj->ops = object_ops;
1853 sgen_workers_enqueue_job (&sj->job, FALSE);
1855 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1856 sj->ops = object_ops;
1857 sgen_workers_enqueue_job (&sj->job, FALSE);
1860 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1863 sgen_pin_stats_print_class_stats ();
1867 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1869 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1870 sgen_finish_pinning ();
1872 sgen_pin_stats_reset ();
1874 if (do_concurrent_checks)
1875 sgen_debug_check_nursery_is_clean ();
1880 major_start_collection (gboolean concurrent, size_t *old_next_pin_slot)
1882 SgenObjectOperations *object_ops;
1884 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1886 current_collection_generation = GENERATION_OLD;
1888 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1891 sgen_cement_reset ();
1894 g_assert (major_collector.is_concurrent);
1895 concurrent_collection_in_progress = TRUE;
1897 object_ops = &major_collector.major_ops_concurrent_start;
1899 object_ops = &major_collector.major_ops_serial;
1902 reset_pinned_from_failed_allocation ();
1904 sgen_memgov_major_collection_start ();
1906 //count_ref_nonref_objs ();
1907 //consistency_check ();
1909 check_scan_starts ();
1912 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1913 gc_stats.major_gc_count ++;
1915 if (major_collector.start_major_collection)
1916 major_collector.start_major_collection ();
1918 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);
1919 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1923 major_finish_collection (const char *reason, size_t old_next_pin_slot, gboolean forced)
1925 ScannedObjectCounts counts;
1926 SgenObjectOperations *object_ops;
1927 mword fragment_total;
1933 if (concurrent_collection_in_progress) {
1934 object_ops = &major_collector.major_ops_concurrent_finish;
1936 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1938 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1940 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1941 main_gc_thread = NULL;
1944 object_ops = &major_collector.major_ops_serial;
1947 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1949 /* all the objects in the heap */
1950 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue));
1952 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1954 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1956 if (objects_pinned) {
1957 g_assert (!concurrent_collection_in_progress);
1960 * This is slow, but we just OOM'd.
1962 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1963 * queue is laid out at this point.
1965 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1967 * We need to reestablish all pinned nursery objects in the pin queue
1968 * because they're needed for fragment creation. Unpinning happens by
1969 * walking the whole queue, so it's not necessary to reestablish where major
1970 * heap block pins are - all we care is that they're still in there
1973 sgen_optimize_pin_queue ();
1974 sgen_find_section_pin_queue_start_end (nursery_section);
1978 reset_heap_boundaries ();
1979 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1981 /* walk the pin_queue, build up the fragment list of free memory, unmark
1982 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1985 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1986 if (!fragment_total)
1988 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1990 if (do_concurrent_checks && concurrent_collection_in_progress)
1991 sgen_debug_check_nursery_is_clean ();
1993 /* prepare the pin queue for the next collection */
1994 sgen_finish_pinning ();
1996 /* Clear TLABs for all threads */
1997 sgen_clear_tlabs ();
1999 sgen_pin_stats_reset ();
2001 sgen_cement_clear_below_threshold ();
2003 if (check_mark_bits_after_major_collection)
2004 sgen_check_heap_marked (concurrent_collection_in_progress);
2007 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2009 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2010 sgen_memgov_major_pre_sweep ();
2013 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2018 time_major_los_sweep += TV_ELAPSED (atv, btv);
2020 major_collector.sweep ();
2022 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2025 time_major_sweep += TV_ELAPSED (btv, atv);
2027 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2029 if (sgen_have_pending_finalizers ()) {
2030 SGEN_LOG (4, "Finalizer-thread wakeup");
2031 sgen_client_finalize_notify ();
2034 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2036 sgen_memgov_major_collection_end (forced);
2037 current_collection_generation = -1;
2039 memset (&counts, 0, sizeof (ScannedObjectCounts));
2040 major_collector.finish_major_collection (&counts);
2042 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
2044 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2045 if (concurrent_collection_in_progress)
2046 concurrent_collection_in_progress = FALSE;
2048 check_scan_starts ();
2050 binary_protocol_flush_buffers (FALSE);
2052 //consistency_check ();
2054 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2058 major_do_collection (const char *reason, gboolean forced)
2060 TV_DECLARE (time_start);
2061 TV_DECLARE (time_end);
2062 size_t old_next_pin_slot;
2064 if (disable_major_collections)
2067 if (major_collector.get_and_reset_num_major_objects_marked) {
2068 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2069 g_assert (!num_marked);
2072 /* world must be stopped already */
2073 TV_GETTIME (time_start);
2075 major_start_collection (FALSE, &old_next_pin_slot);
2076 major_finish_collection (reason, old_next_pin_slot, forced);
2078 TV_GETTIME (time_end);
2079 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2081 /* FIXME: also report this to the user, preferably in gc-end. */
2082 if (major_collector.get_and_reset_num_major_objects_marked)
2083 major_collector.get_and_reset_num_major_objects_marked ();
2085 return bytes_pinned_from_failed_allocation > 0;
2089 major_start_concurrent_collection (const char *reason)
2091 TV_DECLARE (time_start);
2092 TV_DECLARE (time_end);
2093 long long num_objects_marked;
2095 if (disable_major_collections)
2098 TV_GETTIME (time_start);
2099 SGEN_TV_GETTIME (time_major_conc_collection_start);
2101 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2102 g_assert (num_objects_marked == 0);
2104 binary_protocol_concurrent_start ();
2106 // FIXME: store reason and pass it when finishing
2107 major_start_collection (TRUE, NULL);
2109 gray_queue_redirect (&gray_queue);
2111 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2113 TV_GETTIME (time_end);
2114 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2116 current_collection_generation = -1;
2120 * Returns whether the major collection has finished.
2123 major_should_finish_concurrent_collection (void)
2125 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2126 return sgen_workers_all_done ();
2130 major_update_concurrent_collection (void)
2132 TV_DECLARE (total_start);
2133 TV_DECLARE (total_end);
2135 TV_GETTIME (total_start);
2137 binary_protocol_concurrent_update ();
2139 major_collector.update_cardtable_mod_union ();
2140 sgen_los_update_cardtable_mod_union ();
2142 TV_GETTIME (total_end);
2143 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2147 major_finish_concurrent_collection (gboolean forced)
2149 TV_DECLARE (total_start);
2150 TV_DECLARE (total_end);
2152 TV_GETTIME (total_start);
2154 binary_protocol_concurrent_finish ();
2157 * We need to stop all workers since we're updating the cardtable below.
2158 * The workers will be resumed with a finishing pause context to avoid
2159 * additional cardtable and object scanning.
2161 sgen_workers_stop_all_workers ();
2163 SGEN_TV_GETTIME (time_major_conc_collection_end);
2164 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2166 major_collector.update_cardtable_mod_union ();
2167 sgen_los_update_cardtable_mod_union ();
2169 if (mod_union_consistency_check)
2170 sgen_check_mod_union_consistency ();
2172 current_collection_generation = GENERATION_OLD;
2173 sgen_cement_reset ();
2174 major_finish_collection ("finishing", -1, forced);
2176 if (whole_heap_check_before_collection)
2177 sgen_check_whole_heap (FALSE);
2179 TV_GETTIME (total_end);
2180 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2182 current_collection_generation = -1;
2186 * Ensure an allocation request for @size will succeed by freeing enough memory.
2188 * LOCKING: The GC lock MUST be held.
2191 sgen_ensure_free_space (size_t size, int generation)
2193 int generation_to_collect = -1;
2194 const char *reason = NULL;
2196 if (generation == GENERATION_OLD) {
2197 if (sgen_need_major_collection (size)) {
2198 reason = "LOS overflow";
2199 generation_to_collect = GENERATION_OLD;
2202 if (degraded_mode) {
2203 if (sgen_need_major_collection (size)) {
2204 reason = "Degraded mode overflow";
2205 generation_to_collect = GENERATION_OLD;
2207 } else if (sgen_need_major_collection (size)) {
2208 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2209 generation_to_collect = GENERATION_OLD;
2211 generation_to_collect = GENERATION_NURSERY;
2212 reason = "Nursery full";
2216 if (generation_to_collect == -1) {
2217 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2218 generation_to_collect = GENERATION_OLD;
2219 reason = "Finish concurrent collection";
2223 if (generation_to_collect == -1)
2225 sgen_perform_collection (size, generation_to_collect, reason, FALSE);
2229 * LOCKING: Assumes the GC lock is held.
2232 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish)
2234 TV_DECLARE (gc_start);
2235 TV_DECLARE (gc_end);
2236 TV_DECLARE (gc_total_start);
2237 TV_DECLARE (gc_total_end);
2238 GGTimingInfo infos [2];
2239 int overflow_generation_to_collect = -1;
2240 int oldest_generation_collected = generation_to_collect;
2241 const char *overflow_reason = NULL;
2243 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2245 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2247 TV_GETTIME (gc_start);
2249 sgen_stop_world (generation_to_collect);
2251 TV_GETTIME (gc_total_start);
2253 if (concurrent_collection_in_progress) {
2255 * If the concurrent worker is finished or we are asked to do a major collection
2256 * then we finish the concurrent collection.
2258 gboolean finish = major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD;
2261 major_finish_concurrent_collection (wait_to_finish);
2262 oldest_generation_collected = GENERATION_OLD;
2264 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY, "Why aren't we finishing the concurrent collection?");
2265 major_update_concurrent_collection ();
2266 collect_nursery (NULL, FALSE);
2272 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2275 * There's no concurrent collection in progress. Collect the generation we're asked
2276 * to collect. If the major collector is concurrent and we're not forced to wait,
2277 * start a concurrent collection.
2279 // FIXME: extract overflow reason
2280 if (generation_to_collect == GENERATION_NURSERY) {
2281 if (collect_nursery (NULL, FALSE)) {
2282 overflow_generation_to_collect = GENERATION_OLD;
2283 overflow_reason = "Minor overflow";
2286 if (major_collector.is_concurrent && !wait_to_finish) {
2287 collect_nursery (NULL, FALSE);
2288 major_start_concurrent_collection (reason);
2289 // FIXME: set infos[0] properly
2293 if (major_do_collection (reason, wait_to_finish)) {
2294 overflow_generation_to_collect = GENERATION_NURSERY;
2295 overflow_reason = "Excessive pinning";
2299 TV_GETTIME (gc_end);
2301 memset (infos, 0, sizeof (infos));
2302 infos [0].generation = generation_to_collect;
2303 infos [0].reason = reason;
2304 infos [0].is_overflow = FALSE;
2305 infos [1].generation = -1;
2306 infos [0].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2308 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2310 if (overflow_generation_to_collect != -1) {
2312 * We need to do an overflow collection, either because we ran out of memory
2313 * or the nursery is fully pinned.
2316 infos [1].generation = overflow_generation_to_collect;
2317 infos [1].reason = overflow_reason;
2318 infos [1].is_overflow = TRUE;
2321 if (overflow_generation_to_collect == GENERATION_NURSERY)
2322 collect_nursery (NULL, FALSE);
2324 major_do_collection (overflow_reason, wait_to_finish);
2326 TV_GETTIME (gc_end);
2327 infos [1].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2329 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2332 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2334 /* this also sets the proper pointers for the next allocation */
2335 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2336 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2337 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2338 sgen_dump_pin_queue ();
2343 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2345 TV_GETTIME (gc_total_end);
2346 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2348 sgen_restart_world (oldest_generation_collected, infos);
2352 * ######################################################################
2353 * ######## Memory allocation from the OS
2354 * ######################################################################
2355 * This section of code deals with getting memory from the OS and
2356 * allocating memory for GC-internal data structures.
2357 * Internal memory can be handled with a freelist for small objects.
2363 G_GNUC_UNUSED static void
2364 report_internal_mem_usage (void)
2366 printf ("Internal memory usage:\n");
2367 sgen_report_internal_mem_usage ();
2368 printf ("Pinned memory usage:\n");
2369 major_collector.report_pinned_memory_usage ();
2373 * ######################################################################
2374 * ######## Finalization support
2375 * ######################################################################
2379 * If the object has been forwarded it means it's still referenced from a root.
2380 * If it is pinned it's still alive as well.
2381 * A LOS object is only alive if we have pinned it.
2382 * Return TRUE if @obj is ready to be finalized.
2384 static inline gboolean
2385 sgen_is_object_alive (GCObject *object)
2387 if (ptr_in_nursery (object))
2388 return sgen_nursery_is_object_alive (object);
2390 return sgen_major_is_object_alive (object);
2394 * This function returns true if @object is either alive and belongs to the
2395 * current collection - major collections are full heap, so old gen objects
2396 * are never alive during a minor collection.
2399 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2401 if (ptr_in_nursery (object))
2402 return sgen_nursery_is_object_alive (object);
2404 if (current_collection_generation == GENERATION_NURSERY)
2407 return sgen_major_is_object_alive (object);
2412 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2414 return !sgen_is_object_alive (object);
2418 sgen_queue_finalization_entry (GCObject *obj)
2420 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2422 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2424 sgen_client_object_queued_for_finalization (obj);
2428 sgen_object_is_live (GCObject *obj)
2430 return sgen_is_object_alive_and_on_current_collection (obj);
2434 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2435 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2436 * all finalizers have really finished running.
2438 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2439 * This means that just checking whether the queues are empty leaves the possibility that an
2440 * object might have been dequeued but not yet finalized. That's why we need the additional
2441 * flag `pending_unqueued_finalizer`.
2444 static volatile gboolean pending_unqueued_finalizer = FALSE;
2447 sgen_gc_invoke_finalizers (void)
2451 g_assert (!pending_unqueued_finalizer);
2453 /* FIXME: batch to reduce lock contention */
2454 while (sgen_have_pending_finalizers ()) {
2460 * We need to set `pending_unqueued_finalizer` before dequeing the
2461 * finalizable object.
2463 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2464 pending_unqueued_finalizer = TRUE;
2465 mono_memory_write_barrier ();
2466 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2467 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2468 pending_unqueued_finalizer = TRUE;
2469 mono_memory_write_barrier ();
2470 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2476 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2484 /* the object is on the stack so it is pinned */
2485 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2486 sgen_client_run_finalize (obj);
2489 if (pending_unqueued_finalizer) {
2490 mono_memory_write_barrier ();
2491 pending_unqueued_finalizer = FALSE;
2498 sgen_have_pending_finalizers (void)
2500 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2504 * ######################################################################
2505 * ######## registered roots support
2506 * ######################################################################
2510 * We do not coalesce roots.
2513 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2515 RootRecord new_root;
2518 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2519 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2520 /* we allow changing the size and the descriptor (for thread statics etc) */
2522 size_t old_size = root->end_root - start;
2523 root->end_root = start + size;
2524 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2525 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2526 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2527 root->root_desc = descr;
2529 roots_size -= old_size;
2535 new_root.end_root = start + size;
2536 new_root.root_desc = descr;
2537 new_root.source = source;
2540 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2543 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);
2550 sgen_deregister_root (char* addr)
2556 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2557 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2558 roots_size -= (root.end_root - addr);
2564 * ######################################################################
2565 * ######## Thread handling (stop/start code)
2566 * ######################################################################
2570 sgen_get_current_collection_generation (void)
2572 return current_collection_generation;
2576 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2578 #ifndef HAVE_KW_THREAD
2579 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2582 sgen_init_tlab_info (info);
2584 sgen_client_thread_register (info, stack_bottom_fallback);
2590 sgen_thread_unregister (SgenThreadInfo *p)
2592 sgen_client_thread_unregister (p);
2596 * ######################################################################
2597 * ######## Write barriers
2598 * ######################################################################
2602 * Note: the write barriers first do the needed GC work and then do the actual store:
2603 * this way the value is visible to the conservative GC scan after the write barrier
2604 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2605 * the conservative scan, otherwise by the remembered set scan.
2609 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2611 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2612 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2613 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2614 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2618 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2619 if (binary_protocol_is_heavy_enabled ()) {
2621 for (i = 0; i < count; ++i) {
2622 gpointer dest = (gpointer*)dest_ptr + i;
2623 gpointer obj = *((gpointer*)src_ptr + i);
2625 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2630 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2634 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2638 HEAVY_STAT (++stat_wbarrier_generic_store);
2640 sgen_client_wbarrier_generic_nostore_check (ptr);
2642 obj = *(gpointer*)ptr;
2644 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2647 * We need to record old->old pointer locations for the
2648 * concurrent collector.
2650 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2651 SGEN_LOG (8, "Skipping remset at %p", ptr);
2655 SGEN_LOG (8, "Adding remset at %p", ptr);
2657 remset.wbarrier_generic_nostore (ptr);
2661 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2663 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2664 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2665 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2666 mono_gc_wbarrier_generic_nostore (ptr);
2667 sgen_dummy_use (value);
2670 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2671 * as an atomic operation with release semantics.
2674 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2676 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2678 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2680 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2682 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2683 mono_gc_wbarrier_generic_nostore (ptr);
2685 sgen_dummy_use (value);
2689 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2691 GCObject **dest = (GCObject **)_dest;
2692 GCObject **src = (GCObject **)_src;
2696 mono_gc_wbarrier_generic_store (dest, *src);
2701 size -= SIZEOF_VOID_P;
2707 * ######################################################################
2708 * ######## Other mono public interface functions.
2709 * ######################################################################
2713 sgen_gc_collect (int generation)
2718 sgen_perform_collection (0, generation, "user request", TRUE);
2723 sgen_gc_collection_count (int generation)
2725 if (generation == 0)
2726 return gc_stats.minor_gc_count;
2727 return gc_stats.major_gc_count;
2731 sgen_gc_get_used_size (void)
2735 tot = los_memory_usage;
2736 tot += nursery_section->next_data - nursery_section->data;
2737 tot += major_collector.get_used_size ();
2738 /* FIXME: account for pinned objects */
2744 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2748 va_start (ap, description_format);
2750 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2751 vfprintf (stderr, description_format, ap);
2753 fprintf (stderr, " - %s", fallback);
2754 fprintf (stderr, "\n");
2760 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2763 double val = strtod (opt, &endptr);
2764 if (endptr == opt) {
2765 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2768 else if (val < min || val > max) {
2769 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2781 char *major_collector_opt = NULL;
2782 char *minor_collector_opt = NULL;
2783 size_t max_heap = 0;
2784 size_t soft_limit = 0;
2786 gboolean debug_print_allowance = FALSE;
2787 double allowance_ratio = 0, save_target = 0;
2788 gboolean cement_enabled = TRUE;
2791 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2794 /* already inited */
2797 /* being inited by another thread */
2798 mono_thread_info_usleep (1000);
2801 /* we will init it */
2804 g_assert_not_reached ();
2806 } while (result != 0);
2808 SGEN_TV_GETTIME (sgen_init_timestamp);
2810 #ifdef SGEN_WITHOUT_MONO
2811 mono_thread_smr_init ();
2814 mono_coop_mutex_init (&gc_mutex);
2816 gc_debug_file = stderr;
2818 mono_coop_mutex_init (&sgen_interruption_mutex);
2820 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2821 opts = g_strsplit (env, ",", -1);
2822 for (ptr = opts; *ptr; ++ptr) {
2824 if (g_str_has_prefix (opt, "major=")) {
2825 opt = strchr (opt, '=') + 1;
2826 major_collector_opt = g_strdup (opt);
2827 } else if (g_str_has_prefix (opt, "minor=")) {
2828 opt = strchr (opt, '=') + 1;
2829 minor_collector_opt = g_strdup (opt);
2837 sgen_init_internal_allocator ();
2838 sgen_init_nursery_allocator ();
2839 sgen_init_fin_weak_hash ();
2840 sgen_init_hash_table ();
2841 sgen_init_descriptors ();
2842 sgen_init_gray_queues ();
2843 sgen_init_allocator ();
2844 sgen_init_gchandles ();
2846 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2847 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2849 sgen_client_init ();
2851 if (!minor_collector_opt) {
2852 sgen_simple_nursery_init (&sgen_minor_collector);
2854 if (!strcmp (minor_collector_opt, "simple")) {
2856 sgen_simple_nursery_init (&sgen_minor_collector);
2857 } else if (!strcmp (minor_collector_opt, "split")) {
2858 sgen_split_nursery_init (&sgen_minor_collector);
2860 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2861 goto use_simple_nursery;
2865 if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep")) {
2866 use_marksweep_major:
2867 sgen_marksweep_init (&major_collector);
2868 } else if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep-conc")) {
2869 sgen_marksweep_conc_init (&major_collector);
2871 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `marksweep` instead.", "Unknown major collector `%s'.", major_collector_opt);
2872 goto use_marksweep_major;
2875 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2878 gboolean usage_printed = FALSE;
2880 for (ptr = opts; *ptr; ++ptr) {
2882 if (!strcmp (opt, ""))
2884 if (g_str_has_prefix (opt, "major="))
2886 if (g_str_has_prefix (opt, "minor="))
2888 if (g_str_has_prefix (opt, "max-heap-size=")) {
2889 size_t page_size = mono_pagesize ();
2890 size_t max_heap_candidate = 0;
2891 opt = strchr (opt, '=') + 1;
2892 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2893 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2894 if (max_heap != max_heap_candidate)
2895 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2897 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2901 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2902 opt = strchr (opt, '=') + 1;
2903 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2904 if (soft_limit <= 0) {
2905 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2909 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2915 if (g_str_has_prefix (opt, "nursery-size=")) {
2917 opt = strchr (opt, '=') + 1;
2918 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2919 if ((val & (val - 1))) {
2920 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2924 if (val < SGEN_MAX_NURSERY_WASTE) {
2925 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2926 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2930 sgen_nursery_size = val;
2931 sgen_nursery_bits = 0;
2932 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2935 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2941 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2943 opt = strchr (opt, '=') + 1;
2944 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2945 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2950 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2952 opt = strchr (opt, '=') + 1;
2953 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2954 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2955 allowance_ratio = val;
2960 if (!strcmp (opt, "cementing")) {
2961 cement_enabled = TRUE;
2964 if (!strcmp (opt, "no-cementing")) {
2965 cement_enabled = FALSE;
2969 if (!strcmp (opt, "precleaning")) {
2970 precleaning_enabled = TRUE;
2973 if (!strcmp (opt, "no-precleaning")) {
2974 precleaning_enabled = FALSE;
2978 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2981 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2984 if (sgen_client_handle_gc_param (opt))
2987 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2992 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2993 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2994 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2995 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2996 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2997 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2998 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2999 fprintf (stderr, " [no-]cementing\n");
3000 if (major_collector.print_gc_param_usage)
3001 major_collector.print_gc_param_usage ();
3002 if (sgen_minor_collector.print_gc_param_usage)
3003 sgen_minor_collector.print_gc_param_usage ();
3004 sgen_client_print_gc_params_usage ();
3005 fprintf (stderr, " Experimental options:\n");
3006 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3007 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);
3008 fprintf (stderr, "\n");
3010 usage_printed = TRUE;
3015 if (major_collector_opt)
3016 g_free (major_collector_opt);
3018 if (minor_collector_opt)
3019 g_free (minor_collector_opt);
3023 sgen_pinning_init ();
3024 sgen_cement_init (cement_enabled);
3026 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
3027 gboolean usage_printed = FALSE;
3029 opts = g_strsplit (env, ",", -1);
3030 for (ptr = opts; ptr && *ptr; ptr ++) {
3032 if (!strcmp (opt, ""))
3034 if (opt [0] >= '0' && opt [0] <= '9') {
3035 gc_debug_level = atoi (opt);
3040 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3041 gc_debug_file = fopen (rf, "wb");
3043 gc_debug_file = stderr;
3046 } else if (!strcmp (opt, "print-allowance")) {
3047 debug_print_allowance = TRUE;
3048 } else if (!strcmp (opt, "print-pinning")) {
3049 sgen_pin_stats_enable ();
3050 } else if (!strcmp (opt, "verify-before-allocs")) {
3051 verify_before_allocs = 1;
3052 has_per_allocation_action = TRUE;
3053 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3054 char *arg = strchr (opt, '=') + 1;
3055 verify_before_allocs = atoi (arg);
3056 has_per_allocation_action = TRUE;
3057 } else if (!strcmp (opt, "collect-before-allocs")) {
3058 collect_before_allocs = 1;
3059 has_per_allocation_action = TRUE;
3060 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3061 char *arg = strchr (opt, '=') + 1;
3062 has_per_allocation_action = TRUE;
3063 collect_before_allocs = atoi (arg);
3064 } else if (!strcmp (opt, "verify-before-collections")) {
3065 whole_heap_check_before_collection = TRUE;
3066 } else if (!strcmp (opt, "check-at-minor-collections")) {
3067 consistency_check_at_minor_collection = TRUE;
3068 nursery_clear_policy = CLEAR_AT_GC;
3069 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3070 if (!major_collector.is_concurrent) {
3071 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3074 mod_union_consistency_check = TRUE;
3075 } else if (!strcmp (opt, "check-mark-bits")) {
3076 check_mark_bits_after_major_collection = TRUE;
3077 } else if (!strcmp (opt, "check-nursery-pinned")) {
3078 check_nursery_objects_pinned = TRUE;
3079 } else if (!strcmp (opt, "clear-at-gc")) {
3080 nursery_clear_policy = CLEAR_AT_GC;
3081 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3082 nursery_clear_policy = CLEAR_AT_GC;
3083 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3084 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3085 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3086 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3087 } else if (!strcmp (opt, "check-scan-starts")) {
3088 do_scan_starts_check = TRUE;
3089 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3090 do_verify_nursery = TRUE;
3091 } else if (!strcmp (opt, "check-concurrent")) {
3092 if (!major_collector.is_concurrent) {
3093 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3096 nursery_clear_policy = CLEAR_AT_GC;
3097 do_concurrent_checks = TRUE;
3098 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3099 do_dump_nursery_content = TRUE;
3100 } else if (!strcmp (opt, "disable-minor")) {
3101 disable_minor_collections = TRUE;
3102 } else if (!strcmp (opt, "disable-major")) {
3103 disable_major_collections = TRUE;
3104 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3105 char *filename = strchr (opt, '=') + 1;
3106 nursery_clear_policy = CLEAR_AT_GC;
3107 sgen_debug_enable_heap_dump (filename);
3108 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3109 char *filename = strchr (opt, '=') + 1;
3110 char *colon = strrchr (filename, ':');
3113 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3114 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3119 binary_protocol_init (filename, (long long)limit);
3120 } else if (!strcmp (opt, "nursery-canaries")) {
3121 do_verify_nursery = TRUE;
3122 enable_nursery_canaries = TRUE;
3123 } else if (!sgen_client_handle_gc_debug (opt)) {
3124 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3129 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);
3130 fprintf (stderr, "Valid <option>s are:\n");
3131 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3132 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3133 fprintf (stderr, " check-at-minor-collections\n");
3134 fprintf (stderr, " check-mark-bits\n");
3135 fprintf (stderr, " check-nursery-pinned\n");
3136 fprintf (stderr, " verify-before-collections\n");
3137 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3138 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3139 fprintf (stderr, " disable-minor\n");
3140 fprintf (stderr, " disable-major\n");
3141 fprintf (stderr, " check-concurrent\n");
3142 fprintf (stderr, " clear-[nursery-]at-gc\n");
3143 fprintf (stderr, " clear-at-tlab-creation\n");
3144 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3145 fprintf (stderr, " check-scan-starts\n");
3146 fprintf (stderr, " print-allowance\n");
3147 fprintf (stderr, " print-pinning\n");
3148 fprintf (stderr, " heap-dump=<filename>\n");
3149 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3150 fprintf (stderr, " nursery-canaries\n");
3151 sgen_client_print_gc_debug_usage ();
3152 fprintf (stderr, "\n");
3154 usage_printed = TRUE;
3160 if (check_mark_bits_after_major_collection)
3161 nursery_clear_policy = CLEAR_AT_GC;
3163 if (major_collector.post_param_init)
3164 major_collector.post_param_init (&major_collector);
3166 if (major_collector.needs_thread_pool)
3167 sgen_workers_init (1);
3169 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3171 memset (&remset, 0, sizeof (remset));
3173 sgen_card_table_init (&remset);
3175 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");
3181 sgen_get_nursery_clear_policy (void)
3183 return nursery_clear_policy;
3189 mono_coop_mutex_lock (&gc_mutex);
3193 sgen_gc_unlock (void)
3195 gboolean try_free = sgen_try_free_some_memory;
3196 sgen_try_free_some_memory = FALSE;
3197 mono_coop_mutex_unlock (&gc_mutex);
3199 mono_thread_hazardous_try_free_some ();
3203 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3205 major_collector.iterate_live_block_ranges (callback);
3209 sgen_get_major_collector (void)
3211 return &major_collector;
3215 sgen_get_remset (void)
3221 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3223 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3224 sgen_los_count_cards (los_total, los_marked);
3227 static gboolean world_is_stopped = FALSE;
3229 /* LOCKING: assumes the GC lock is held */
3231 sgen_stop_world (int generation)
3233 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3235 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3237 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3239 sgen_client_stop_world (generation);
3241 world_is_stopped = TRUE;
3243 if (binary_protocol_is_heavy_enabled ())
3244 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3245 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3248 /* LOCKING: assumes the GC lock is held */
3250 sgen_restart_world (int generation, GGTimingInfo *timing)
3252 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3254 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3256 if (binary_protocol_is_heavy_enabled ())
3257 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3258 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3260 sgen_client_restart_world (generation, timing);
3262 world_is_stopped = FALSE;
3264 binary_protocol_world_restarted (generation, sgen_timestamp ());
3266 sgen_try_free_some_memory = TRUE;
3268 if (sgen_client_bridge_need_processing ())
3269 sgen_client_bridge_processing_finish (generation);
3271 sgen_memgov_collection_end (generation, timing, timing ? 2 : 0);
3275 sgen_is_world_stopped (void)
3277 return world_is_stopped;
3281 sgen_check_whole_heap_stw (void)
3283 sgen_stop_world (0);
3284 sgen_clear_nursery_fragments ();
3285 sgen_check_whole_heap (FALSE);
3286 sgen_restart_world (0, NULL);
3290 sgen_timestamp (void)
3292 SGEN_TV_DECLARE (timestamp);
3293 SGEN_TV_GETTIME (timestamp);
3294 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3297 #endif /* HAVE_SGEN_GC */